[{"k":"docs","t":"What Graspable is","x":"Graspable is a desktop app where you describe a 3D or XR experience and an AI agent builds it in a project on your computer. Graspable is a desktop app for Windows, macOS and Linux. You describe a 3D or XR experience in plain language. An AI agent edits a real project folder on your computer, checks that the result builds and runs, and shows it in a live preview. You can then try it on a supported","s":"Graspable is a desktop app where you describe a 3D or XR experience and an AI agent builds it in a project on your computer. Graspable is a desktop app for Windows, macOS and Linux. You describe a 3D or XR experience in plain language. An AI agent edits a real project folder on your computer, checks that the result builds and runs, and shows it in a live preview. You can then try it on a supported headset or phone. Web projects use Vite, React and React Three Fiber. Native templates use Unity, Unreal Engine or Swift for visionOS and need their engine tools installed. You can read and edit your project files in Graspable or another editor. See Templates and Native XR for requirements.","u":"/docs"},{"k":"docs","t":"What Graspable is","h":"How a run works","x":"1. You type a request, for example \"add a red ball that bounces when I tap it\". You can also say it, or talk with the agent and hear its answers. See Voice. 2. The agent reads the project and edits files. The editor updates as it writes. 3. Risky actions, such as running a command, ask you first. 4. For web projects, Graspable builds the project, loads it in a browser on your computer and replays ","s":"1. You type a request, for example \"add a red ball that bounces when I tap it\". You can also say it, or talk with the agent and hear its answers. See Voice. 2. The agent reads the project and edits files. The editor updates as it writes. 3. Risky actions, such as running a command, ask you first. 4. For web projects, Graspable builds the project, loads it in a browser on your computer and replays your configured XR recordings. Native projects use their engine's checks. 5. If a check fails, the agent gets up to three repair turns. The run completes only when the checks pass.","u":"/docs#how-a-run-works"},{"k":"docs","t":"What Graspable is","h":"What you need","x":"- A Windows x64 PC, a Mac (Apple silicon or Intel), or a Linux x64 computer. - One way to power the agent: a ChatGPT subscription, a Claude subscription, or an API key from a model provider. A local model also works. See Engines and sign-in. - Web projects use the runtime included with Graspable. The browser that checks your scenes is found on your computer or downloaded once. Native projects need","s":"- A Windows x64 PC, a Mac (Apple silicon or Intel), or a Linux x64 computer. - One way to power the agent: a ChatGPT subscription, a Claude subscription, or an API key from a model provider. A local model also works. See Engines and sign-in. - Web projects use the runtime included with Graspable. The browser that checks your scenes is found on your computer or downloaded once. Native projects need their engine tools. See Install. - A headset is optional. The preview includes an emulated one. A Graspable account is optional. You need it only for plans and for sending ratings and anonymized run summaries.","u":"/docs#what-you-need"},{"k":"docs","t":"What Graspable is","h":"Where to go next","x":"- Install Graspable - Your first project - Engines and sign-in - Preview and headset - Voice - Troubleshooting Questions? Ask on Discord.","u":"/docs#where-to-go-next"},{"k":"docs","t":"Install Graspable","x":"Install Graspable on Windows, macOS or Linux, and find your projects and settings. ","u":"/docs/install"},{"k":"docs","t":"Install Graspable","h":"Before you install","x":"For web projects, Graspable includes the runtime and manages the checking browser: - What runs your projects is built into the app. It installs a project's packages, builds it and runs its preview. You do not need Node.js. - What checks that scenes load is a browser. On a Mac, Graspable downloads Chrome for Testing from Google once, about 190 MB, the first time the app starts, and uses your own Ch","s":"For web projects, Graspable includes the runtime and manages the checking browser: - What runs your projects is built into the app. It installs a project's packages, builds it and runs its preview. You do not need Node.js. - What checks that scenes load is a browser. On a Mac, Graspable downloads Chrome for Testing from Google once, about 190 MB, the first time the app starts, and uses your own Chrome until that is done: Chrome for Testing never updates itself, so macOS does not ask about Graspable changing an app. On Windows and Linux it uses Chrome, Edge, Chromium or Brave if you have one, and downloads Chrome for Testing only if you have none. You can start building while it downloads. On Linux, Graspable unpacks the download itself; you do not need a separate archive tool. Both are listed in Settings → Components . Under Advanced you can choose the Node.js on your computer or your own Chromium-based browser instead. Native projects need their engine tools, such as Unity, Unreal Engine or Xcode. See Native XR for platform requirements.","u":"/docs/install#before-you-install"},{"k":"docs","t":"Install Graspable","h":"Download","x":"Get the installer for your system from graspable.dev/download. Choose Older versions beside the system information below the installers to browse previous releases. System Installer --- --- macOS, Apple silicon .dmg macOS, Intel .dmg Windows, x64 .exe installer Linux, x64 .AppImage , .deb or .rpm Every installer is published with its SHA-256 checksum and a signed checksum file. The download page s","s":"Get the installer for your system from graspable.dev/download. Choose Older versions beside the system information below the installers to browse previous releases. System Installer --- --- macOS, Apple silicon .dmg macOS, Intel .dmg Windows, x64 .exe installer Linux, x64 .AppImage , .deb or .rpm Every installer is published with its SHA-256 checksum and a signed checksum file. The download page shows the checksums and the public key. Choose Copy checksum beside an installer to copy its SHA-256 value. If your browser blocks clipboard access, a selectable checksum appears so you can copy it manually. On a phone, use Email yourself the link on the download page and open the email on your Windows, Mac or Linux computer. The editor runs on your computer; published experiences can be opened on phones when their browser supports the features used. The news checkbox is optional.","u":"/docs/install#download"},{"k":"docs","t":"Install Graspable","h":"First launch","x":"On macOS, open the .dmg and drag Graspable to Applications . The app is signed and checked by Apple, so it opens like any other app. On Windows, open the .exe installer and follow its steps. On Linux, use the package installer for the .deb or .rpm , or mark the AppImage executable in its file properties and open it. Graspable keeps sign-ins and API keys in your system's keychain. On macOS, the sys","s":"On macOS, open the .dmg and drag Graspable to Applications . The app is signed and checked by Apple, so it opens like any other app. On Windows, open the .exe installer and follow its steps. On Linux, use the package installer for the .deb or .rpm , or mark the AppImage executable in its file properties and open it. Graspable keeps sign-ins and API keys in your system's keychain. On macOS, the system may ask for your login password when the app first reads the keychain. When the app opens you see the Projects page. The top right shows \"connecting to agent…\" for a moment while the agent starts.","u":"/docs/install#first-launch"},{"k":"docs","t":"Install Graspable","h":"Updating","x":"Graspable looks for a new version when it starts and every few hours while it is open. When there is one, an Update button appears in the bottom left corner. Click it: the app downloads the update, installs it and restarts. The next desktop patch downloads the update while you work, then waits for code and animation edits to be saved and current work to finish before installing. If it asks you to ","s":"Graspable looks for a new version when it starts and every few hours while it is open. When there is one, an Update button appears in the bottom left corner. Click it: the app downloads the update, installs it and restarts. The next desktop patch downloads the update while you work, then waits for code and animation edits to be saved and current work to finish before installing. If it asks you to save first, retrying uses the downloaded package. Editing is briefly locked during installation and restart. You can also look yourself: open Settings → Updates and choose Check for updates . The app verifies each update against a signing key built into it before installing.","u":"/docs/install#updating"},{"k":"docs","t":"Install Graspable","h":"Where your data lives","x":"What Where --- --- Projects ~/Graspable/projects/<id — one normal folder per project. Choose another folder under Settings → Projects folder ; the run history and backups move with it. Run history .runs inside the projects folder Backups of projects moved to the cloud .cloud inside the projects folder Memory about you ~/.graspable/memory Memory about a project .graspable/memory inside the project ","s":"What Where --- --- Projects ~/Graspable/projects/<id — one normal folder per project. Choose another folder under Settings → Projects folder ; the run history and backups move with it. Run history .runs inside the projects folder Backups of projects moved to the cloud .cloud inside the projects folder Memory about you ~/.graspable/memory Memory about a project .graspable/memory inside the project folder ChatGPT sign-in for the Graspable agent ~/.graspable/agent Certificate for headset sharing ~/.graspable/headset List of your cloud projects ~/.graspable/cloud.json API keys, your Graspable account sign-in, cloud provider keys your system's keychain ~ is your home folder.","u":"/docs/install#where-your-data-lives"},{"k":"docs","t":"Install Graspable","h":"Uninstalling","x":"Remove the app the way you remove any app on your system. This does not delete your projects or settings. To remove everything, also delete the ~/Graspable and ~/.graspable folders. Deleting ~/Graspable deletes your projects, so copy out anything you want to keep first. On macOS, the keychain item is named vault under dev.graspable.desktop ; you can remove it in Keychain Access. Claude Code and th","s":"Remove the app the way you remove any app on your system. This does not delete your projects or settings. To remove everything, also delete the ~/Graspable and ~/.graspable folders. Deleting ~/Graspable deletes your projects, so copy out anything you want to keep first. On macOS, the keychain item is named vault under dev.graspable.desktop ; you can remove it in Keychain Access. Claude Code and the Codex CLI keep their own sign-ins in their own places. Uninstalling Graspable does not touch them.","u":"/docs/install#uninstalling"},{"k":"docs","t":"Your first project","x":"Create a project from a template, send your first request, read the run, open the preview and undo a run. Before you start, choose what powers the agent in Settings → AI for the agent . See Engines and sign-in.","u":"/docs/first-project"},{"k":"docs","t":"Your first project","h":"Create a project","x":"1. On the Projects page, type a Name . 2. Pick a Template . See Templates. 3. Choose Create . Graspable copies the template into a new folder and installs its packages. The project header shows \"installing packages\" while this runs. You can look at the files and write your first request in the meantime: the agent starts as soon as the packages are in. The preview unlocks when the install ends.","u":"/docs/first-project#create-a-project"},{"k":"docs","t":"Your first project","h":"Send the first request","x":"The project page has three parts: the file list on the left, the editor in the middle, and the run panel on the right. Type what you want in the box that says \"What should we build?\" and press Enter. Shift+Enter adds a new line. You can also say it. Press the microphone beside the box, speak, and press it again. What you said appears in the box, where you can change it before you send it. Voice ne","s":"The project page has three parts: the file list on the left, the editor in the middle, and the run panel on the right. Type what you want in the box that says \"What should we build?\" and press Enter. Shift+Enter adds a new line. You can also say it. Press the microphone beside the box, speak, and press it again. What you said appears in the box, where you can change it before you send it. Voice needs a Graspable account or your own ElevenLabs key. See Voice. Say what should happen, not how to code it. For example: While the agent works, the box says \"Steer the agent…\". Anything you send then reaches the agent during the same run. The square button cancels the run.","u":"/docs/first-project#send-the-first-request"},{"k":"docs","t":"Your first project","h":"Read the run","x":"The run panel shows the run as it happens. You see It means --- --- Text The agent talking to you Thinking… The agent's reasoning as it streams: the latest lines, with the earlier ones fading out above. A finished thought folds to one line, Thought · 120 words ; click it to read the whole text. How much the Graspable agent reasons is set under Settings → AI for the agent → Reasoning ▸ name … The a","s":"The run panel shows the run as it happens. You see It means --- --- Text The agent talking to you Thinking… The agent's reasoning as it streams: the latest lines, with the earlier ones fading out above. A finished thought folds to one line, Thought · 120 words ; click it to read the whole text. How much the Graspable agent reasons is set under Settings → AI for the agent → Reasoning ▸ name … The agent used a tool, such as reading a file or running a command ✎ App.jsx The agent wrote that file. The editor shows the new content, keeping any unsaved edits for review 🧠 recalled: … Items from memory were given to the agent for this request An amber card with Allow and Deny The agent wants to do something that needs your approval. See Approvals and safety ✓ build check or ✗ build check A check passed or failed — success — The run completed","u":"/docs/first-project#read-the-run"},{"k":"docs","t":"Your first project","h":"The checks","x":"When the agent finishes editing, Graspable checks the result: 1. build — the project is built. 2. smoke — the built app is loaded in a browser on your computer, without a window. Errors, failed requests and a blank page fail the check. A screenshot of what the browser saw appears under the check. 3. xr — your XR recordings are replayed. This check only appears when the project has recordings. See ","s":"When the agent finishes editing, Graspable checks the result: 1. build — the project is built. 2. smoke — the built app is loaded in a browser on your computer, without a window. Errors, failed requests and a blank page fail the check. A screenshot of what the browser saw appears under the check. 3. xr — your XR recordings are replayed. This check only appears when the project has recordings. See XR tests. If a check fails, the agent gets the error and tries to fix it. It gets at most three repair turns. If the check still fails, the run ends as failed and tells you which check. If no browser is found, the smoke and xr checks are skipped and the check line gives the reason.","u":"/docs/first-project#the-checks"},{"k":"docs","t":"Your first project","h":"Open the preview","x":"Use the buttons at the top right of the project page: - Code — the editor only. - Split — the editor above the preview. - Preview — the preview only. The first click starts the project's dev server. - Compare — runs the same request with several engines side by side. You can then Adopt one result or Discard it. The preview reloads as files change. See Preview and headset.","u":"/docs/first-project#open-the-preview"},{"k":"docs","t":"Your first project","h":"Edit by hand","x":"Click a file to open it. Type, then choose Save , or press ⌘S on macOS and Ctrl+S on Windows and Linux. Version 0.19.1 labels the button with your platform's shortcut. Hold Alt (Option on macOS) and drag a decimal number sideways to change it, or hold Shift too for smaller steps. The preview updates as you drag. From version 0.19.1, numbers can become draggable beyond the initially parsed part of ","s":"Click a file to open it. Type, then choose Save , or press ⌘S on macOS and Ctrl+S on Windows and Linux. Version 0.19.1 labels the button with your platform's shortcut. Hold Alt (Option on macOS) and drag a decimal number sideways to change it, or hold Shift too for smaller steps. The preview updates as you drag. From version 0.19.1, numbers can become draggable beyond the initially parsed part of a large file, and saved code updates appear without an extra typing delay when you have no unsaved edits. Version 0.19.1 also keeps your unsaved code when the agent, project history or another editor changes the file. View saved version lets you review the new text; Replace saved version asks before saving your edits over it. Discard edits returns to the saved text. Before a new agent request or comparison, Graspable saves your code drafts. If a save fails or a file changed elsewhere, your request waits and your edits remain. Navigating away from a project with unsaved code asks first. Drafts are held while the project is open; save before quitting the app. The next desktop patch asks before closing or quitting with unsaved code, unsaved animation, or work in progress in the open project. On macOS, this includes Command-Q and Quit in the app menu. Choose Keep editing to return without stopping the agent. This does not recover work after a crash or forced quit. From version 0.19.1, Undo and Redo stay separate for each file. Graspable remembers the cursor and scroll position in your 12 most recently used files while the project stays open. Switching between Code, Split and Preview keeps this history. If a saved file changes while you are away from it, its old editing history is cleared when you reopen it. If you select code in the editor, a chip above the prompt box shows the file and lines. The selection goes with your next request, so you can write \"make this slower\" without naming the file. Click the × on the chip to leave it out. From version 0.19.1, the selected text and line numbers also update as you edit, so your next request receives the current selection.","u":"/docs/first-project#edit-by-hand"},{"k":"docs","t":"Your first project","h":"Assets","x":"Beside Files , the Assets tab lists the project's models, images, sounds, videos and the engine's own assets (a Unity prefab, an Unreal Blueprint or level), by folder. Type to find one, or show one kind only. Version 0.19.1 loads image thumbnails as you scroll and releases those that leave view, keeping large asset folders lighter. A selected image stays open until you close its preview. Search ig","s":"Beside Files , the Assets tab lists the project's models, images, sounds, videos and the engine's own assets (a Unity prefab, an Unreal Blueprint or level), by folder. Type to find one, or show one kind only. Version 0.19.1 loads image thumbnails as you scroll and releases those that leave view, keeping large asset folders lighter. A selected image stays open until you close its preview. Search ignores letter case and keeps your selected preview open while you filter the list. If nothing matches, Clear filters restores all assets. If the list could not load, choose Retry . Click an asset to see it. The preview scrolls within the sidebar so the asset list remains reachable; Close returns to the list. A model (glTF, GLB, OBJ, STL, 3MF and PCD point clouds) is drawn in 3D with its textures: drag to turn it, scroll to move closer. Triangle counts help you compare exports; they do not measure headset frame rate. Images, sound and video are shown and played. Other kinds are listed with their size. Add copies files from your computer into the project, each into the folder its kind belongs in: public/models , public/textures , public/audio and public/video in a web project; Assets/Models and so on in Unity; Source/Models and so on in an Unreal project; App/Resources/ in a Vision Pro app. The agent uses the same folders, so you can add a model and then ask for it by its file name. If an asset uses symbolic links, use Add to copy the original files into the project before previewing them. Version 0.19.1 applies the same protected-file checks to previews as other asset reads. Linked source files inside the project still work, but links cannot be used to change Graspable’s project settings, saved history or recorded headset tests. You can bring existing models from Meshy, Tripo, Luma Genie or another tool. For web projects, export a GLB with its skeleton and animation clips included. Selecting a GLB or glTF in Assets shows its skins, joints, morph targets and clip names. Choose a clip, play or pause it, or scrub to inspect a pose. Morph controls lets you test the model's existing facial expressions and deformations, with each mesh named beside its controls. Fit model brings the current pose into view without changing your viewing direction. For animated models, Restore pose restores the imported pose and expressions without moving the camera. It becomes available after you change the pose. Focus the preview and use Shift plus arrow keys to orbit, or arrow keys to pan. These are audition controls: they do not write changes to the model or save a performance into your animation take. A model without facial controls does not gain them from this preview. Draco, Meshopt and KTX2-compressed assets use bundled decoders. Preview loading is cancelled when you leave the model, and rendering stops when idle or hidden. Large exports may need smaller textures: previews allow up to 192 MB per file, 256 MB in total, 64 files, 20,000 scene nodes, 512 MB of estimated geometry and animation attributes, a separate 512 MB limit on decoded model buffers, and 64 million texture pixels, with a maximum texture edge of 8,192 pixels. The morph panel shows up to 256 controls and reports when there are more. A glTF import copies its local buffers and textures together. Missing or external resources, unsupported required features and decoding failures stop the preview instead of silently dropping parts of the model. Existing filenames are preserved by giving a new import a numbered name. See Characters and models for connectors, rigging and provider-library limits. Closing a GLB or glTF preview releases resources from every scene in the model, including scenes that were not displayed. If decoding finishes after you close it, that result is released too. An interrupted graphics preview stops and offers Retry . The texture limit covers every decoded scene, including textures on line and point geometry. A texture shared between scenes is counted once. Closing the preview releases line and point geometry as well as triangle meshes. Version 0.19.1 checks glTF object hierarchies before decoding. Cycles, repeated parent links and missing child objects produce an error; hierarchies deeper than 256 levels need unnecessary groups flattened before previewing. These checks leave the original file intact. STL, 3MF and PCD previews also avoid creating an unnecessary text copy of the model. OBJ, STL, 3MF and PCD previews redraw when you move or resize the view, then rest while idle. Closing or switching previews cancels their pending downloads. The same patch rejects malformed ZIP64 data in archive-based models such as 3MF, instead of leaving the preview unresponsive. The original file stays untouched; re-export it from its source application if the preview reports an invalid archive. That patch also refreshes media previews when an asset's modification time changes and releases their temporary data when no thumbnail or preview needs it. A failed media download offers Retry in the preview. From version 0.17.2, USD assets have an Open USD inspector action for scene hierarchy, properties, variants and separate edit layers. Build preview converts supported data to a web model; the original USD remains available. GLB and glTF assets also offer Export as USD . See OpenUSD scenes for supported formats and conversion limits.","u":"/docs/first-project#assets"},{"k":"docs","t":"Your first project","h":"Go back in time and undo","x":"When a run has more than one event, a slider appears at the top of the run panel. Drag it back to see the transcript and the files as they were at that point. The editor is read-only while you do this. Live returns to the present. When a run that changed files has ended, you can undo it: - Undo run puts every file the run changed back to how it was before the run. - With the slider dragged back, t","s":"When a run has more than one event, a slider appears at the top of the run panel. Drag it back to see the transcript and the files as they were at that point. The editor is read-only while you do this. Live returns to the present. When a run that changed files has ended, you can undo it: - Undo run puts every file the run changed back to how it was before the run. - With the slider dragged back, the button reads Undo after here and puts the files back to how they were at that point. If you changed a file after the run, Graspable leaves it alone and tells you how many files \"changed since\". Choose overwrite them too to undo those as well. Undo never deletes a file that existed before the run started.","u":"/docs/first-project#go-back-in-time-and-undo"},{"k":"docs","t":"Your first project","h":"History","x":"Undo works on one run. History , in the project header, is the whole story of the project: a list of versions, newest first. A version is a snapshot of every file in the project at one moment, with a short note. Graspable records one when the project is created (\"Start from the … template\") and one at the end of every run that changed files — the note is your request. A run that failed or was stop","s":"Undo works on one run. History , in the project header, is the whole story of the project: a list of versions, newest first. A version is a snapshot of every file in the project at one moment, with a short note. Graspable records one when the project is created (\"Start from the … template\") and one at the end of every run that changed files — the note is your request. A run that failed or was stopped is recorded too, so nothing it did is lost. Click a version to see which files it added, changed or removed. Restore this version puts every file back to how it was then, and asks first. Restoring does not delete anything from the history: it adds a new version (\"Restore to …\") on top, so you can always come forward again. Installed packages, the preview and the agent's memory are not part of versions and are left alone. Projects made before this feature get their history the next time a run ends or the list is opened; the first version then is the project as it was at that moment. Keep the history on your GitHub On the Pro plan and higher, a project's versions can also go to a repository on your own GitHub. Choose GitHub in the project header, then Connect GitHub : a GitHub page opens in your browser and asks for the short code shown in Graspable. Back in the app, give the repository a name (it is made if it does not exist, private unless you untick that) and choose Link and push . From then on every run's version is pushed there by itself. Push now does it by hand, and Unlink stops it; the repository on GitHub stays as it is. Graspable only ever adds to a repository: if it has commits that are not in the project, the push is refused and the panel says so. The history travels with the project when it moves to a Cloud Agent and back. Each version is a git commit, so a project folder also works with git tools you may already use; nothing needs to be installed for Graspable itself. The same list is available as grasp history in the CLI and project history in the MCP server.","u":"/docs/first-project#history"},{"k":"docs","t":"Templates","x":"The project templates that ship with Graspable, what each starts you with, and how a project is laid out. A template is the starting point of a project. You pick it once, when you create the project. Graspable copies it into your project folder and installs its packages. It also gives the agent instructions that fit the template, so a short request goes a long way. On the Projects screen, choose B","s":"The project templates that ship with Graspable, what each starts you with, and how a project is laid out. A template is the starting point of a project. You pick it once, when you create the project. Graspable copies it into your project folder and installs its packages. It also gives the agent instructions that fit the template, so a short request goes a long way. On the Projects screen, choose Browse templates . Compare the pictures and descriptions, or search for what you want to make. Open Details to see requirements and requests to try, then choose Use this template (or Select template for a plan you do not have yet). Choosing a template does not create the project: give it a name and choose Create . If the online catalog is unavailable, the app still lists the templates included with it; pictures need a connection. Twenty-two templates ship with the app; each has a page under Templates, with requests to try and what people made from it. React XR is for everyone. The others are premium templates: the WebXR and Expo ones come with the Builder plan and higher, the native ones (Unity, Unreal Engine, Apple Vision Pro) with the Pro plan and higher; see Plans and billing. To create a project from a premium template you must be signed in, so the app can check your plan. A project you already created keeps working whatever your plan. Template Kind Use it for --- --- --- Tabletop AR Premium, WebAR A rescue film and relay puzzle on a real table, with phone image tracking and WebXR surface placement Unity AR tabletop Premium, native (Unity 6) The same miniature world in a native iOS, Android or Quest app React XR Free, WebXR Anything built from objects, starting from a small working scene Splat World Premium, WebXR Standing inside a generated place and adding things to it World Explorer Premium, WebXR Walking through a generated place as if it were solid: tours, viewings, exhibitions World Game Premium, WebXR A game inside a generated place, with physics WebGPU Splats Premium, WebXR Millions of Gaussian splats on WebGPU: captured places and splats made by code Scanned Room Premium, WebXR Measuring, planning and training inside a capture of a real room Product Studio Premium, WebXR Product viewers and configurators, placed at real size XR Game Premium, WebXR Games you play by grabbing, throwing and hitting Hands & Spatial UI Premium, WebXR Apps you use with your bare hands: panels, dashboards, instruments Multi-user Premium, WebXR A room several people are in at once: meetings, games, reviews Cross-platform XR Premium, WebXR One app that adapts to every headset, pair of glasses, phone and computer, in your own room where the device shows it Panel app Premium, Expo A flat app that opens as a window in a headset's home space, and as a web page everywhere else Meta glasses app Premium, Expo A phone app for Ray-Ban Meta, Oakley Meta and Meta Ray-Ban Display glasses: camera, display, inputs, motion, speech Glasses field notes Premium, Expo A hands-free photo journal for Ray-Ban Meta and Oakley Meta glasses: look, tap the temple, say a note Display glasses web app Premium A web app that runs on Meta Ray-Ban Display glasses, installed from a link Unity for Quest Premium, native (Unity 6) A native mixed-reality app for Meta Quest and Meta VR Glasses: passthrough, real surfaces, hands, controllers Unity for Android XR Premium, native (Unity 6) The same kit for Android XR headsets and glasses Unity for XREAL Premium, native (Unity 6) The same kit for XREAL glasses Apple Vision Pro Premium, native (Swift) A visionOS app: a window and an immersive space, built with Xcode Unreal Engine XR Premium, native (Unreal Engine 5) A scene for OpenXR headsets whose level is a Python script your Unreal Editor builds The native templates need an engine on your computer; see Native XR.","u":"/docs/templates"},{"k":"docs","t":"Templates","h":"React XR","x":"See the React XR template. The project starts with a bench in front of you: a cube, a can and a ball to pick up and throw, a New ball and a Tidy up button, a sign, and a basket that counts what lands in it. It works with controllers, tracked hands and a mouse, and has an Enter VR and an Enter AR button; in AR the bench stands in your own room. Each part is a small file the agent reuses: things you","s":"See the React XR template. The project starts with a bench in front of you: a cube, a can and a ball to pick up and throw, a New ball and a Tidy up button, a sign, and a basket that counts what lands in it. It works with controllers, tracked hands and a mouse, and has an Enter VR and an Enter AR button; in AR the bench stands in your own room. Each part is a small file the agent reuses: things you can pick up, solid boxes, buttons, text, sound. It uses React Three Fiber for the 3D scene and @react-three/xr for headsets. The physics library @react-three/rapier and the helper library @react-three/drei are installed too. With tracked hands, a thumbs-up adds a ball, like the New ball button. In AR, the floor, walls, tables and furniture your headset finds in the room become solid: a ball you throw bounces off your real table and rolls across your real floor. See Gestures and the real room.","u":"/docs/templates#react-xr"},{"k":"docs","t":"Templates","h":"Tabletop AR and Unity AR tabletop","x":"Tabletop AR starts with Signal Garden : a 90-second rescue film and a puzzle where you turn three relays to guide a courier home. It includes animated models, physically based materials, original Gaussian terrain and a musical score. Watch it from the film camera or move around the miniature yourself. Pause, seek, mute and hide the controls; the scene remains editable. In a compatible WebXR browse","s":"Tabletop AR starts with Signal Garden : a 90-second rescue film and a puzzle where you turn three relays to guide a courier home. It includes animated models, physically based materials, original Gaussian terrain and a musical score. Watch it from the film camera or move around the miniature yourself. Pause, seek, mute and hide the controls; the scene remains editable. In a compatible WebXR browser, place the garden on a detected surface. Other phone browsers use the included mission mat: print it or show it on a second screen, then point the camera at it. The board disappears and playback pauses when the mat leaves view. This is image tracking, so it needs the mat throughout the experience. Unity AR tabletop contains the same film, puzzle and assets. It has separate phone AR and Quest configurations. The phone version uses ARKit on iOS and ARCore on Android; the Quest version uses OpenXR passthrough. Choose the target in Unity before building. Graspable exports the Android build without automatically installing it on a connected device. See AR films and tabletop games for device requirements and the editing workflow.","u":"/docs/templates#tabletop-ar-and-unity-ar-tabletop"},{"k":"docs","t":"Templates","h":"The three world templates","x":"Splat World World Explorer World Game Splat World, World Explorer and World Game start inside a world : a place generated by World Labs and shown as Gaussian splats with the Spark renderer. To get a different place, ask for it, for example \"make this a greenhouse at dusk\". See Generated worlds. - Splat World starts in a sunlit loft. You look around, teleport and put things in it. They sit on the w","s":"Splat World World Explorer World Game Splat World, World Explorer and World Game start inside a world : a place generated by World Labs and shown as Gaussian splats with the Spark renderer. To get a different place, ask for it, for example \"make this a greenhouse at dusk\". See Generated worlds. - Splat World starts in a sunlit loft. You look around, teleport and put things in it. They sit on the world's real floor and tables, at real size. - World Explorer starts in the same loft, but solid: floors hold you, walls and furniture stop you. It comes with info points and a guided tour. On a flat screen you walk with W A S D; in the headset with the thumbstick or by teleporting. - World Game starts in a tavern with a throwing game: knock the mugs off the bar, ring the targets, keep score. Balls bounce off the room itself. A world is a folder in the project, public/worlds/<name / . Switching to another world is one word in the code, and the agent does it for you after generating one.","u":"/docs/templates#the-three-world-templates"},{"k":"docs","t":"Templates","h":"WebGPU Splats","x":"WebGPU Splats Gaussian splats drawn with Radix, the WebGPU splat renderer from the makers of Spark. Every splat in the scene is projected, sorted and drawn together on the graphics card, so a recent computer shows one to two million of them smoothly. The project starts with a spiral galaxy of 600 000 splats that code makes on the graphics card every frame. - Splats from a file: put a .spz or .ply ","s":"WebGPU Splats Gaussian splats drawn with Radix, the WebGPU splat renderer from the makers of Spark. Every splat in the scene is projected, sorted and drawn together on the graphics card, so a recent computer shows one to two million of them smoothly. The project starts with a spiral galaxy of 600 000 splats that code makes on the graphics card every frame. - Splats from a file: put a .spz or .ply capture in public/splats/ and ask for it. Radix does not read .splat , .ksplat or .sog files. - Splats made by code: a generator is a short shader that makes each splat from its number: galaxies, fractals, particle fountains, flowers that open as you come near. Ask for what you want to see; the agent writes the generator. - In a headset: WebGPU draws on computers and phones, and in a headset whose browser can draw a headset session with WebGPU: on a Quest, switch on the WebGPU and WebXR experiments in the browser's chrome://flags and restart it. Otherwise, and in the preview's emulated headset, the same scene runs on Radix's WebGL2 path, which handles a few hundred thousand splats at a headset's frame rate. Add ?gpu=webgpu-gl or ?gpu=webgpu to the address to choose one. This template has no physics engine. For physics, start from React XR or XR Game.","u":"/docs/templates#webgpu-splats"},{"k":"docs","t":"Templates","h":"Scanned Room","x":"See the Scanned Room template. The project starts inside a 3D capture of a real seating corner, at real scale in metres. A reticle lies on the real floor, walls and furniture where you point; you can move around the room and place objects that sit on those surfaces. Use it for measuring tools, furniture planning, guided tours and training on site. You can replace the room with your own scan. The t","s":"See the Scanned Room template. The project starts inside a 3D capture of a real seating corner, at real scale in metres. A reticle lies on the real floor, walls and furniture where you point; you can move around the room and place objects that sit on those surfaces. Use it for measuring tools, furniture planning, guided tours and training on site. You can replace the room with your own scan. The template says what the file needs: metres, y up, the floor at height 0. The room that ships with the template is a capture from the Eyeful Tower dataset, copyright (c) Meta Platforms, Inc. and affiliates, under the MIT License. The licence text is in the project beside the file.","u":"/docs/templates#scanned-room"},{"k":"docs","t":"Templates","h":"Product Studio","x":"See the Product Studio template. The project starts with a chair on a photo-studio stage: soft light, reflections and a contact shadow. A panel lets people choose colours and materials, see an exploded view, read callouts, and place the product at its real size in VR or in their own room. To show your own product, give the agent its 3D file and say what its parts are.","u":"/docs/templates#product-studio"},{"k":"docs","t":"Templates","h":"XR Game","x":"See the XR Game template. The project starts in an arcade booth with a can-knockdown game: pick up a ball, throw it, score. Physics, grabbing and throwing, a scoreboard, a timer and sounds are ready, so a request like \"turn it into basketball\" is a small change. With tracked hands, a thumbs-up after a game starts the next one, like the START OVER button.","u":"/docs/templates#xr-game"},{"k":"docs","t":"Templates","h":"Hands & Spatial UI","x":"See the Hands & Spatial UI template. The project starts with a floating panel of sliders, buttons and a toggle that drive an object in front of you. Everything works by pinching and poking with your hands, with controllers, and with a mouse. Use it for dashboards, data in space, instruments and tools.","u":"/docs/templates#hands-amp-spatial-ui"},{"k":"docs","t":"Templates","h":"Multi-user","x":"See the Multi-user template. The project starts as a room that several people open at the same time. Everyone sees the others as avatars with a name plate, a head and two hands, can pick up the same objects on the table and press the same board on the wall, and all of it stays in step. It works in a headset and on a flat screen together. In the preview, everyone who opens the same preview address ","s":"See the Multi-user template. The project starts as a room that several people open at the same time. Everyone sees the others as avatars with a name plate, a head and two hands, can pick up the same objects on the table and press the same board on the wall, and all of it stays in step. It works in a headset and on a flat screen together. In the preview, everyone who opens the same preview address is in the same room. Once you publish the app, everyone who opens it meets in the app's own room; add ?room=name to the address for a separate one. Up to 16 people share a room, and what was changed stays for a day after the last person leaves. The shared state travels through Graspable's relay; nothing else leaves the app. You can ask the agent to make any of your own objects shared: \"make the score shared between everyone\".","u":"/docs/templates#multi-user"},{"k":"docs","t":"Templates","h":"Cross-platform XR","x":"See the Cross-platform XR template. One WebXR app for every device. Where the device has passthrough (Meta Quest, Android XR, phones) the app is seen in your own room, placed on real surfaces. Where it has not (Apple Vision Pro's browser, a computer) the same scene stands in a photographed room. The start scene is a lounge chair with a side table that you can place, turn and refinish, beside a pan","s":"See the Cross-platform XR template. One WebXR app for every device. Where the device has passthrough (Meta Quest, Android XR, phones) the app is seen in your own room, placed on real surfaces. Where it has not (Apple Vision Pro's browser, a computer) the same scene stands in a photographed room. The start scene is a lounge chair with a side table that you can place, turn and refinish, beside a panel with three tabs: the product, live data fetched from the internet, and what the app found out about the device it runs on. It looks real: the furniture models and the room's light are photographed (Poly Haven, public domain), and every surface is a physically based material. The panel is built with uikit, so it is laid out like a web page and used with controllers, bare hands, look-and-pinch, or a mouse. On glasses with a narrow view the panel becomes a small card near the middle. A published app from this template can be installed from the browser, and its one address serves every device.","u":"/docs/templates#cross-platform-xr"},{"k":"docs","t":"Templates","h":"Panel app","x":"See the Panel app template. Not every headset app is a 3D scene. Meta Quest, Meta VR Glasses and Android XR run ordinary Android apps as floating panels in the home space, and a panel is often the right shape for a tool: a list, a dashboard, a reader. This template is an Expo project (React Native) that starts as a to-do list laid out for a panel, with big targets you point at from a distance. It ","s":"See the Panel app template. Not every headset app is a 3D scene. Meta Quest, Meta VR Glasses and Android XR run ordinary Android apps as floating panels in the home space, and a panel is often the right shape for a tool: a list, a dashboard, a reader. This template is an Expo project (React Native) that starts as a to-do list laid out for a panel, with big targets you point at from a distance. It runs as a web page in the preview and anywhere else, and builds as an Android app for the headsets with Expo's own tools. Because it is plain JavaScript, it works with a Cloud Agent too.","u":"/docs/templates#panel-app"},{"k":"docs","t":"Templates","h":"Meta glasses app","x":"See the Meta glasses app template. A phone app for Meta AI glasses: Ray-Ban Meta Gen 1 and Gen 2, Ray-Ban Meta Optics, Oakley Meta HSTN and Vanguard, and Meta Ray-Ban Display. It is an Expo project on the Meta Wearables Device Access Toolkit and shows everything the toolkit can do, one screen each: registering with the Meta AI app, the camera stream and photos, a screen on Display glasses, the gla","s":"See the Meta glasses app template. A phone app for Meta AI glasses: Ray-Ban Meta Gen 1 and Gen 2, Ray-Ban Meta Optics, Oakley Meta HSTN and Vanguard, and Meta Ray-Ban Display. It is an Expo project on the Meta Wearables Device Access Toolkit and shows everything the toolkit can do, one screen each: registering with the Meta AI app, the camera stream and photos, a screen on Display glasses, the glasses' inputs, motion and speech, launching by voice, and mock glasses for development without hardware. The preview runs it against the glasses emulator; your own glasses need the app on a phone. See Meta glasses.","u":"/docs/templates#meta-glasses-app"},{"k":"docs","t":"Templates","h":"Glasses field notes","x":"See the Glasses field notes template. A finished-looking app for glasses without a display: Ray-Ban Meta Gen 1 and Gen 2 and Oakley Meta. You look through the glasses on your phone in a lens-shaped viewfinder, save a moment with the shutter or a tap on the temple, say a note and it is written under the photo, and hear \"saved\" in the glasses' own speakers. A line across the view levels the horizon ","s":"See the Glasses field notes template. A finished-looking app for glasses without a display: Ray-Ban Meta Gen 1 and Gen 2 and Oakley Meta. You look through the glasses on your phone in a lens-shaped viewfinder, save a moment with the shutter or a tap on the temple, say a note and it is written under the photo, and hear \"saved\" in the glasses' own speakers. A line across the view levels the horizon from the glasses' motion sensor. Moments stay on the phone. Start here when you want an app people would use; start from the Meta glasses app when you want to see every capability on its own. See Meta glasses.","u":"/docs/templates#glasses-field-notes"},{"k":"docs","t":"Templates","h":"Display glasses web app","x":"See the Display glasses web app template. A web app for Meta Ray-Ban Display glasses, built with Meta's UI Toolkit: 600 by 600 pixels, moved through with the Neural Band or the touchpad, no pointer. It starts with a timer, guided steps, the glasses' senses and a note by voice. Publish it and the glasses install it from a code. See Meta glasses.","u":"/docs/templates#display-glasses-web-app"},{"k":"docs","t":"Templates","h":"Unity for Quest, Android XR and XREAL","x":"Unity for Quest Unity for Android XR Unity for XREAL Three native Unity 6 projects with the same Graspable kit and a different headset target each. The start scene is mixed reality: the real room shows through, a small table holds a ball you can grab and a button you can press, and you place a marker on any real surface by pointing and pinching. The kit has scripts for labels at a readable size, b","s":"Unity for Quest Unity for Android XR Unity for XREAL Three native Unity 6 projects with the same Graspable kit and a different headset target each. The start scene is mixed reality: the real room shows through, a small table holds a ball you can grab and a button you can press, and you place a marker on any real surface by pointing and pinching. The kit has scripts for labels at a readable size, buttons, grabbable things, placing on real surfaces, and a hint that follows you. See Native XR.","u":"/docs/templates#unity-for-quest-android-xr-and-xreal"},{"k":"docs","t":"Templates","h":"Apple Vision Pro","x":"See the Apple Vision Pro template. A native visionOS app in Swift: a window with controls and an immersive space around you, in which a shape spins and can be pinched and dragged. SwiftUI for the window, RealityKit for the space, built and run in the visionOS Simulator by your own Xcode. See Native XR.","u":"/docs/templates#apple-vision-pro"},{"k":"docs","t":"Templates","h":"Unreal Engine XR","x":"See the Unreal Engine XR template. An Unreal Engine 5 project whose level is written as a Python script and built by your own Unreal Editor, so the agent can change it and you can read what changed. It starts with a workbench on a concrete floor in photographed materials. See Native XR for what it does and does not do yet.","u":"/docs/templates#unreal-engine-xr"},{"k":"docs","t":"Templates","h":"Gestures and the real room","x":"The React XR and XR Game templates recognise gestures, so you can ask the agent for things like \"a thumbs-up starts the next round\" or \"nod to confirm\". - Hand gestures , with tracked hands: pinch, point, fist, thumbs-up, open palm and spread fingers. Each one is reported when it starts, while it is held and when it ends, for the left and the right hand. - Head gestures , in a headset: a nod and a","s":"The React XR and XR Game templates recognise gestures, so you can ask the agent for things like \"a thumbs-up starts the next round\" or \"nod to confirm\". - Hand gestures , with tracked hands: pinch, point, fist, thumbs-up, open palm and spread fingers. Each one is reported when it starts, while it is held and when it ends, for the left and the right hand. - Head gestures , in a headset: a nod and a shake of the head. Controllers and flat screens make no gestures, so keep a button for anything a gesture does. In the React XR template, AR also uses the room you are in. The surfaces your headset detects (floor, walls, table tops) and the room shapes it scans become solid, so things land on your real table and bounce off your real walls. On Quest 3 and Android XR headsets, real things can also hide virtual ones behind them, using the headset's depth sensing. That costs some performance, so it is off until you ask for it: ask the agent to \"turn on occlusion\". The preview in Graspable shows the scene without occlusion.","u":"/docs/templates#gestures-and-the-real-room"},{"k":"docs","t":"Templates","h":"How a web project is laid out","x":"Every web template has the same layout. There is no src folder. The premium templates add ready-made parts under components/ and lib/ , and files such as worlds, a room or a product under public/ . File What it is --- --- App.jsx The root of your app. Most changes start here App.css Styles components/ Where the agent puts additional parts of the scene index.jsx , index.html The entry point package","s":"Every web template has the same layout. There is no src folder. The premium templates add ready-made parts under components/ and lib/ , and files such as worlds, a room or a product under public/ . File What it is --- --- App.jsx The root of your app. Most changes start here App.css Styles components/ Where the agent puts additional parts of the scene index.jsx , index.html The entry point package.json , vite.config.js Packages and build settings tests/xr/ Your XR recordings. See XR tests The Panel app template is an Expo project: App.tsx is the root, app.json the app's settings. The native templates are laid out as their engine expects; Native XR has the table.","u":"/docs/templates#how-a-web-project-is-laid-out"},{"k":"docs","t":"AR films and tabletop games","x":"Put an editable miniature film or game on a real table, in a phone browser or a native AR app. Choose Tabletop AR for a browser experience, or Unity AR tabletop for a native phone or Quest app. Both start with Signal Garden: a caretaker, a stranded garden and a courier that needs a working landing beacon. Watch film plays the 90-second rescue. Play mission lets you turn the three relays until sunl","s":"Put an editable miniature film or game on a real table, in a phone browser or a native AR app. Choose Tabletop AR for a browser experience, or Unity AR tabletop for a native phone or Quest app. Both start with Signal Garden: a caretaker, a stranded garden and a courier that needs a working landing beacon. Watch film plays the 90-second rescue. Play mission lets you turn the three relays until sunlight reaches the landing pad. The channels show the route, Hint identifies the next relay, and Undo takes back a turn. You can change the story, puzzle, objects and animation by asking the agent or editing the project yourself.","u":"/docs/augmented-reality"},{"k":"docs","t":"AR films and tabletop games","h":"Try Signal Garden","x":"Open Signal Garden to play the relay puzzle or watch the film. You can try both on a flat screen without an account or camera access. For iPhone AR, first open the mission mat on another screen or print it at 24 cm wide. Lay it flat, choose AR with a mat in the demo, then open the camera and keep the whole picture in view. The browser tracks that picture; it cannot detect a bare table. For a phone","s":"Open Signal Garden to play the relay puzzle or watch the film. You can try both on a flat screen without an account or camera access. For iPhone AR, first open the mission mat on another screen or print it at 24 cm wide. Lay it flat, choose AR with a mat in the demo, then open the camera and keep the whole picture in view. The browser tracks that picture; it cannot detect a bare table. For a phone app that detects surfaces without a mat, use the native Unity template. On a Quest or Android phone with a compatible WebXR browser, open the same demo and choose Enter AR . Placement depends on the device's surface support, as described below.","u":"/docs/augmented-reality#try-signal-garden"},{"k":"docs","t":"AR films and tabletop games","h":"Put it on your table","x":"Device How placement works --- --- Android phone with an AR-capable WebXR browser Enter AR, aim at a horizontal surface, then tap to place Quest with a compatible browser Enter AR, find a surface, then select with a controller or hand iPhone Safari, or a browser without surface AR Open the mission mat from Settings on another screen or print it; enter AR and point the phone at it Native iOS app AR","s":"Device How placement works --- --- Android phone with an AR-capable WebXR browser Enter AR, aim at a horizontal surface, then tap to place Quest with a compatible browser Enter AR, find a surface, then select with a controller or hand iPhone Safari, or a browser without surface AR Open the mission mat from Settings on another screen or print it; enter AR and point the phone at it Native iOS app ARKit surface tracking; built and signed through Unity and Xcode Native Android phone app ARCore surface tracking on supported phones Native Quest app OpenXR passthrough; select the Quest target in Unity The web experience needs a secure connection and camera permission. On browsers without surface AR, choose AR with a mat . The setup shows the picture and explains how to print it or display it on another screen before you open the camera. Keep the whole mat in view: losing it hides the scene and pauses the film; finding it resumes at the same moment. If surface detection is unavailable in an immersive session, the scene explicitly offers placement in front of you. That fallback does not detect a physical table. These are separate tracking paths. A device that supports one does not necessarily support the others. WebAR on iPhone uses the mat; it does not offer markerless surface tracking.","u":"/docs/augmented-reality#put-it-on-your-table"},{"k":"docs","t":"AR films and tabletop games","h":"Watch or play","x":"On a flat screen, drag to inspect the miniature. In film mode, Settings → Follow film camera switches between the authored shots and your own view. In AR, your movement always controls the viewpoint. The film never moves your head camera. In WebXR, if no horizontal surface is found after five seconds, a placement ring appears in front of you. Look where you want the garden, then pinch or press the","s":"On a flat screen, drag to inspect the miniature. In film mode, Settings → Follow film camera switches between the authored shots and your own view. In AR, your movement always controls the viewpoint. The film never moves your head camera. In WebXR, if no horizontal surface is found after five seconds, a placement ring appears in front of you. Look where you want the garden, then pinch or press the trigger. This places it in your tracked space without claiming a detected table. Leave AR stays available in the floating controls, including before placement and when the main controls are hidden. Touch a floating button with your index fingertip, or point and pinch to select it from a distance. Buttons highlight when targeted. In the rescue game, the same gestures turn the relays. Settings also contains size and sound controls. Hide clears the main controls, and Show controls brings them back. The Made with Graspable link stays visible. Headsets have a small panel beside the garden for playback and the puzzle. Native phone controls stay inside the screen’s safe area when you rotate the phone; hiding them also hides the scene captions.","u":"/docs/augmented-reality#watch-or-play"},{"k":"docs","t":"AR films and tabletop games","h":"Build your own AR game with AI","x":"1. Install Graspable, create a project and choose Tabletop AR for the web version, or Unity AR tabletop for a native build. Choose the web template if you want to share a browser link and can use a mat on iPhone. 2. Start with a change you can check in the preview. For example: “Rename Orchard to Greenhouse. Keep its relay connected to the same channels.” Ask the agent to make the change, then tur","s":"1. Install Graspable, create a project and choose Tabletop AR for the web version, or Unity AR tabletop for a native build. Choose the web template if you want to share a browser link and can use a mat on iPhone. 2. Start with a change you can check in the preview. For example: “Rename Orchard to Greenhouse. Keep its relay connected to the same channels.” Ask the agent to make the change, then turn that relay and use Undo to check that the puzzle still works. 3. Change the story, models or animation. Use the animation workspace to edit the film manually, or describe the change to your agent. Try the film and puzzle separately after changing either one. 4. For a web project, publish and share its link. Open that link on each target device and check placement, a relay turn, Undo, film playback, sound and leaving AR. For a native project, follow the build requirements below. Further requests you can try: - “Turn the outpost into a lunar greenhouse. Make the robot inspect each plant before calling the courier.” - “Add a fourth relay and let me undo any turn.” - “Replace the outpost with my model. Keep the garden sixty centimetres across.” - “Change the final shot to follow the courier, but keep the AR viewpoint under the viewer’s control.” The film take contains editable object transforms, camera shots and score timing. The robot's walk follows the distance it travels, so changing the path does not make its feet slide. Imported models retain their materials and animation clips. The included rocky terrain is an original synthetic Gaussian scene, not a scan of a real place.","u":"/docs/augmented-reality#build-your-own-ar-game-with-ai"},{"k":"docs","t":"AR films and tabletop games","h":"Native builds","x":"The Unity template needs Unity 6 and the build modules for your target. In Unity, choose Signal Garden → Target → Phone AR or Quest , then build. Build Android app in Graspable exports the selected Android configuration; it does not choose a device or install it automatically. For iOS, use Unity's iOS module and Xcode signing. Camera access must be allowed on the phone. The native film activates i","s":"The Unity template needs Unity 6 and the build modules for your target. In Unity, choose Signal Garden → Target → Phone AR or Quest , then build. Build Android app in Graspable exports the selected Android configuration; it does not choose a device or install it automatically. For iOS, use Unity's iOS module and Xcode signing. Camera access must be allowed on the phone. The native film activates iOS media playback when sound is on; use the phone’s media volume to set loudness. Phone builds target 60 frames per second, with actual performance depending on the device and scene. The native Gaussian renderer is intended for the small included terrain. Large captures need a separate performance budget and device testing; replacing this terrain with a much larger file is not a performance guarantee. The Quest target uses Android 12L or newer; the phone target keeps its separate Android requirement. Select the target before building so the correct tracking and platform settings are applied.","u":"/docs/augmented-reality#native-builds"},{"k":"docs","t":"AR films and tabletop games","h":"Plans and costs","x":"Tabletop AR comes with Builder and higher plans; native Unity templates come with Pro and higher. Viewing and local tracking do not consume Credits. Agent work and paid generation use the existing model or generation charges shown by Graspable. The web template self-hosts the open-source 8th Wall image tracker. It does not include the separately licensed 8th Wall world-tracking binary or a paid tr","s":"Tabletop AR comes with Builder and higher plans; native Unity templates come with Pro and higher. Viewing and local tracking do not consume Credits. Agent work and paid generation use the existing model or generation charges shown by Graspable. The web template self-hosts the open-source 8th Wall image tracker. It does not include the separately licensed 8th Wall world-tracking binary or a paid tracking service. Unity's own licensing terms still apply to native development.","u":"/docs/augmented-reality#plans-and-costs"},{"k":"docs","t":"Preview and headset","x":"Use the live preview, the Scene panel, the emulated headset and number scrubbing, and open your project on a real headset. ","u":"/docs/preview-and-headset"},{"k":"docs","t":"Preview and headset","h":"The preview","x":"Choose Preview or Split at the top right of a project. Graspable starts the project's own dev server on your computer and shows it in the app. The preview updates as you or the agent change files. You can return to Code or Compare while it starts; finishing startup will not switch you away from that view. For web projects, the round-arrow button reloads the preview. It also restarts the dev server","s":"Choose Preview or Split at the top right of a project. Graspable starts the project's own dev server on your computer and shows it in the app. The preview updates as you or the agent change files. You can return to Code or Compare while it starts; finishing startup will not switch you away from that view. For web projects, the round-arrow button reloads the preview. It also restarts the dev server if it has stopped. Repeated requests while a preview starts share the same server. Stopping the preview cancels startup too, so it cannot reopen in the background. Native projects use the capture or simulator control inside their preview pane instead of a web reload. Starting an agent check leaves an unchanged preview configuration alone, avoiding an unnecessary restart of the preview you are using. Each preview gets an available local address. Graspable checks both IPv4 and IPv6 loopback ports so another preview does not prevent it from starting. An address already in use during either check is retried automatically. If an address becomes occupied after those checks, during startup, the error remains visible; use the round-arrow button to try a new address. On Windows and Linux, the ⌥ key in this page is the Alt key.","u":"/docs/preview-and-headset#the-preview"},{"k":"docs","t":"Preview and headset","h":"The Scene panel","x":"Choose Scene in the bar above the preview. A panel lists every object in the running scene, as a tree. - Click a row to outline that object in the preview and select the code that created it. - Type in filter objects… to narrow the list. - Hidden objects are dimmed. - Tick rig to also show controllers, hands and pointers. - Each row shows the file and line the object comes from. Objects that your ","s":"Choose Scene in the bar above the preview. A panel lists every object in the running scene, as a tree. - Click a row to outline that object in the preview and select the code that created it. - Type in filter objects… to narrow the list. - Hidden objects are dimmed. - Tick rig to also show controllers, hands and pointers. - Each row shows the file and line the object comes from. Objects that your project's code did not write have none.","u":"/docs/preview-and-headset#the-scene-panel"},{"k":"docs","t":"Preview and headset","h":"From an object to its code","x":"Hold ⌥ and click an object in the preview. The editor opens the file and selects the code that created the object. The matching row is selected in the Scene panel. Version 0.19.1 fixes selection when the object comes from a different file: the editor waits for that file before applying its source range. Invalid source coordinates are ignored instead of interrupting editing. That selection goes wit","s":"Hold ⌥ and click an object in the preview. The editor opens the file and selects the code that created the object. The matching row is selected in the Scene panel. Version 0.19.1 fixes selection when the object comes from a different file: the editor waits for that file before applying its source range. Invalid source coordinates are ignored instead of interrupting editing. That selection goes with your next request, shown as a chip above the prompt box. So you can ⌥-click a planet and write \"make this twice as big\". It works the other way too. Move the cursor in a .jsx or .tsx file and the objects written on that line are outlined in the preview.","u":"/docs/preview-and-headset#from-an-object-to-its-code"},{"k":"docs","t":"Preview and headset","h":"Scrub a number","x":"In the editor, hold ⌥ over a number and drag left or right. The number changes and the preview follows while you drag. Hold Shift for steps ten times finer. The file is saved when you let go. Use Split to see the code and the preview together.","u":"/docs/preview-and-headset#scrub-a-number"},{"k":"docs","t":"Preview and headset","h":"The emulated headset","x":"The preview includes an emulated headset, so you can try AR and VR without one. Choose which in the bar above the preview: Quest 3 (controllers and hands), or Meta VR Glasses (aim with your eyes, pinch to select, a narrower view, no controllers). Changing it reloads the preview. 1. Click Enter AR or Enter VR in the preview. 2. Use the on-screen controls to move the headset and controllers and pres","s":"The preview includes an emulated headset, so you can try AR and VR without one. Choose which in the bar above the preview: Quest 3 (controllers and hands), or Meta VR Glasses (aim with your eyes, pinch to select, a narrower view, no controllers). Changing it reloads the preview. 1. Click Enter AR or Enter VR in the preview. 2. Use the on-screen controls to move the headset and controllers and press their buttons with mouse and keyboard. For AR, the emulated headset stands in a synthetic room, so placing things on floors and walls works. Pick the room from the menu in the bar above the preview: living room, meeting room, music room, office large or office small. In AR or VR, a Talk to the agent button sits low in front of you. Press it to talk with the agent while you look at the scene: it uses this computer's microphone and reads the agent's answers aloud. See Voice. The emulated headset exists only in the preview. It is left out when the project is built.","u":"/docs/preview-and-headset#the-emulated-headset"},{"k":"docs","t":"Preview and headset","h":"Captures and the project's picture","x":"Capture , in the bar above the preview, takes a picture of what you see in the preview at that moment, from where you stand and look. It is always the same shape, 1280 × 720, whatever the size of the pane: the middle of the view is kept. It is the preview's own frame, not a picture of your screen, and it is instant. The picture opens before anything is kept. From there you can: - Use as the projec","s":"Capture , in the bar above the preview, takes a picture of what you see in the preview at that moment, from where you stand and look. It is always the same shape, 1280 × 720, whatever the size of the pane: the middle of the view is kept. It is the preview's own frame, not a picture of your screen, and it is instant. The picture opens before anything is kept. From there you can: - Use as the project's picture. This is the picture lists show for the project and your published app shows on the site. It can be changed once in ten seconds; a published app shows the new one the next time you publish. - Keep in Captures , to have it later. - Download it as a file. - Take it again , after moving to a better spot in the preview. - Close , and nothing is kept. Until you choose one, the project's picture is taken by itself: it is the app as a person first sees it, from the check the agent runs when it finishes a request. The Captures tab, beside Files and Assets, shows the project's picture and every capture you kept. Open one to download it, make it the project's picture, remove it, or share it: the app writes a short post with the project's name, its address when it is published, and \"via graspable.dev\", which you can copy or open in X or LinkedIn. Attach the downloaded picture to the post. Captures are kept in the project on your computer and are sent nowhere. They are not part of the project's files: the agent does not see them and they are not published with the source. A project made before version 0.13 says its preview cannot take pictures yet and offers Update it , which brings the preview's tools up to date.","u":"/docs/preview-and-headset#captures-and-the-project-39-s-picture"},{"k":"docs","t":"Preview and headset","h":"Record and replay","x":"1. Enter AR or VR in the preview. 2. Choose Record , then use your app. 3. Choose Stop & save . The recording is saved in the project as a file in tests/xr/ . Choose replay fixture… to play one back. After the replay, the bar says whether the scene and the final image match the recording. Recordings also work as tests. See XR tests.","u":"/docs/preview-and-headset#record-and-replay"},{"k":"docs","t":"Preview and headset","h":"Open on a real headset","x":"Choose Headset in the bar above the preview. Edits keep hot-reloading on the headset. The headset uses its own WebXR, not the emulated one. Headset preview is not available for cloud projects. Move the project back to this computer first. Same Wi-Fi 1. Choose Share on Wi-Fi . 2. Scan the QR code with the headset, or choose Copy link and open the link in the headset's browser. 3. The first time, th","s":"Choose Headset in the bar above the preview. Edits keep hot-reloading on the headset. The headset uses its own WebXR, not the emulated one. Headset preview is not available for cloud projects. Move the project back to this computer first. Same Wi-Fi 1. Choose Share on Wi-Fi . 2. Scan the QR code with the headset, or choose Copy link and open the link in the headset's browser. 3. The first time, the headset browser warns that the connection is not private. This is expected: the certificate is made on your computer. Choose Advanced , then Proceed . 4. Choose Stop sharing when you are done. Anyone on your network who has the link sees the preview until you stop sharing. Only the preview is shared. Sharing also stops when the preview stops. If your computer has more than one network address, a menu under the QR code lets you pick another. To talk with the agent while you wear the headset, choose Talk in the headset under the QR code and scan the new code. In VR or AR, press Talk to the agent and speak. The headset listens with its own microphone and plays the agent's answers. See Voice. USB (Quest, Pico, Android XR) USB avoids the certificate warning. It needs two things that Graspable does not install: - Android platform-tools ( adb ) on your computer. - Developer mode turned on in the headset. Then: 1. Plug the headset in with USB. 2. Put the headset on and allow USB debugging for this computer. 3. In the Headset panel, choose Open next to the headset. The preview opens in the headset's browser. If the preview does not open on your headset, see Troubleshooting.","u":"/docs/preview-and-headset#open-on-a-real-headset"},{"k":"docs","t":"Generated worlds","x":"Ask for a place and get it as a 3D world you can stand in, generated by World Labs and saved into your project as Gaussian splats. A world is a place: a loft, a tavern, a forest clearing. Ask for one in plain words and the agent generates it with World Labs and saves it into your project. You can then walk through it, put things in it and build on it. Worlds are Gaussian splats: a way to show a pl","s":"Ask for a place and get it as a 3D world you can stand in, generated by World Labs and saved into your project as Gaussian splats. A world is a place: a loft, a tavern, a forest clearing. Ask for one in plain words and the agent generates it with World Labs and saves it into your project. You can then walk through it, put things in it and build on it. Worlds are Gaussian splats: a way to show a place as it looks, with real light and detail, instead of as modelled geometry. Graspable shows them with the Spark renderer.","u":"/docs/worlds"},{"k":"docs","t":"Generated worlds","h":"Ask for a world","x":"In a project made from one of the splat templates, just ask: Make this a greenhouse workshop at dusk, full of plants and glassware, with warm light from hanging lamps. The agent asks you before it starts, because a world costs Credits. Generation takes about five minutes. The run waits for it and then switches the scene to the new world. Good descriptions say what kind of place it is, what is in i","s":"In a project made from one of the splat templates, just ask: Make this a greenhouse workshop at dusk, full of plants and glassware, with warm light from hanging lamps. The agent asks you before it starts, because a world costs Credits. Generation takes about five minutes. The run waits for it and then switches the scene to the new world. Good descriptions say what kind of place it is, what is in it, the light and the mood. One or two sentences are enough.","u":"/docs/worlds#ask-for-a-world"},{"k":"docs","t":"Generated worlds","h":"Sizes","x":"Size What you get Credits --- --- --- Draft A quick, rough version to check the idea 2 Standard The default 11 Large A bigger outdoor place or a large interior 21 A world that fails to generate costs nothing. Generating worlds through Graspable comes with the Builder plan and higher. See Credits.","u":"/docs/worlds#sizes"},{"k":"docs","t":"Generated worlds","h":"With your own World Labs account","x":"If you have a World Labs API key, you can use it instead of Credits. Open Settings → Plugins and connectors , switch on World Labs and enter the key. World Labs then bills you directly, and the Graspable plan does not matter.","u":"/docs/worlds#with-your-own-world-labs-account"},{"k":"docs","t":"Generated worlds","h":"What is saved","x":"Each world is a folder in your project, public/worlds/<name / : File What it is --- --- world-500k.spz The world, as 500,000 splats. This is what is shown world-100k.spz A light version collider.glb The same place as a plain mesh, for standing on floors and bumping into walls. It is never shown pano.png The panorama the world was made from world.json Its name, description, the scale that turns it ","s":"Each world is a folder in your project, public/worlds/<name / : File What it is --- --- world-500k.spz The world, as 500,000 splats. This is what is shown world-100k.spz A light version collider.glb The same place as a plain mesh, for standing on floors and bumping into walls. It is never shown pano.png The panorama the world was made from world.json Its name, description, the scale that turns it into metres, and the credit line The templates read world.json for you, so a world appears at real size with its floor at your feet. A world you generated is yours to keep in the project. Worlds are generated with World Labs Marble; keep the credit in world.json with the files.","u":"/docs/worlds#what-is-saved"},{"k":"docs","t":"Generated worlds","h":"Use a splat of your own","x":"A world does not have to be generated. A Gaussian splat you captured yourself, or made in another tool, works in the same three templates. 1. Open a project made from Splat World, World Explorer or World Game. 2. Choose Worlds in the project header, then Choose a file… and pick the splat: .spz , .ply , .splat , .ksplat or .sog , up to 500 MB. 3. Choose Use it next to the new world. The agent switc","s":"A world does not have to be generated. A Gaussian splat you captured yourself, or made in another tool, works in the same three templates. 1. Open a project made from Splat World, World Explorer or World Game. 2. Choose Worlds in the project header, then Choose a file… and pick the splat: .spz , .ply , .splat , .ksplat or .sog , up to 500 MB. 3. Choose Use it next to the new world. The agent switches the app to it. The file is copied into the project as public/worlds/<name / . It stays on your computer; nothing is uploaded until you publish the app. Things to know: - Size and floor. A splat does not say how big it is or where its floor is. If the place looks too big or too small, or the floor is not under your feet, tell the agent, for example \"the room should be about 5 metres wide\" or \"the floor is a metre too low\". It corrects the two numbers in the world's world.json . - Upside down? Most splat files have their y axis pointing down, and Graspable expects that. Tick its y axis points up when you add one that comes in upside down. - Nothing is solid. A splat of your own has no collider mesh, so the floor is one flat plane at your feet: you can walk and things can land on it, but nothing rests on the furniture you see. - Keep it small. A published app may be 60 to 500 MB depending on your plan, a project on a Cloud Agent 100 MB to 1 GB, and a headset has to download all of it. A .ply saved as .spz is about ten times smaller with no visible loss; one to two million splats is plenty for a room.","u":"/docs/worlds#use-a-splat-of-your-own"},{"k":"docs","t":"Generated worlds","h":"The templates that use worlds","x":"- Splat World : look around a world and add things to it. - World Explorer : walk through a world as if it were solid. - World Game : a game inside a world, with physics. See Templates.","u":"/docs/worlds#the-templates-that-use-worlds"},{"k":"docs","t":"XR tests","x":"Recordings of you using your app in XR become tests that Graspable replays after every run to catch things the agent broke. An XR test is a recording of someone using the app in AR or VR. Graspable replays it after each run and compares the result with what was recorded. This catches changes that break something you already had working.","u":"/docs/xr-tests"},{"k":"docs","t":"XR tests","h":"What a recording holds","x":"A recording is one file in the project's tests/xr/ folder, with a name ending in .xrrec.json . It holds: - every movement and button press of the headset and controllers, - whether the session was AR or VR, and which synthetic room it was in, - the scene at the end: each object's type, name, position and colour, - a picture of the final frame. The files are part of the project, so they travel with","s":"A recording is one file in the project's tests/xr/ folder, with a name ending in .xrrec.json . It holds: - every movement and button press of the headset and controllers, - whether the session was AR or VR, and which synthetic room it was in, - the scene at the end: each object's type, name, position and colour, - a picture of the final frame. The files are part of the project, so they travel with it.","u":"/docs/xr-tests#what-a-recording-holds"},{"k":"docs","t":"XR tests","h":"Make a recording","x":"1. Open the preview and click Enter AR or Enter VR . 2. Choose Record in the bar above the preview. 3. Use the app the way you want it to keep working. 4. Choose Stop & save . See Preview and headset for the emulated headset. The agent can also add recordings. It has a tool that opens the app in a browser, performs scripted actions and saves them as a recording. These show up as file edits in the ","s":"1. Open the preview and click Enter AR or Enter VR . 2. Choose Record in the bar above the preview. 3. Use the app the way you want it to keep working. 4. Choose Stop & save . See Preview and headset for the emulated headset. The agent can also add recordings. It has a tool that opens the app in a browser, performs scripted actions and saves them as a recording. These show up as file edits in the run. The agent's tool never overwrites a recording you made, and re-recording one of its own asks you first.","u":"/docs/xr-tests#make-a-recording"},{"k":"docs","t":"XR tests","h":"How the completion check uses them","x":"After the build and smoke checks pass, Graspable replays every recording in tests/xr/ , in a browser without a window. The app enters XR through its own Enter AR or Enter VR button, which is why that button must stay in the app. There are three outcomes. Outcome What you see What happens --- --- --- Scenes match ✓ xr check — … replayed, scenes match The run can complete An error during replay ✗ xr","s":"After the build and smoke checks pass, Graspable replays every recording in tests/xr/ , in a browser without a window. The app enters XR through its own Enter AR or Enter VR button, which is why that button must stay in the app. There are three outcomes. Outcome What you see What happens --- --- --- Scenes match ✓ xr check — … replayed, scenes match The run can complete An error during replay ✗ xr check — … failed The agent gets a repair turn, like a failed build The scene is different ✗ xr check — … differ from the recording Nothing is repaired. You decide The check compares objects by type, name, position and colour. Invisible objects are ignored. Text drawn into a texture is not seen by the check. If an object's state matters, ask the agent to give the object a meaningful name. If the project has no recordings, there is no xr check.","u":"/docs/xr-tests#how-the-completion-check-uses-them"},{"k":"docs","t":"XR tests","h":"Accept a new baseline","x":"A different scene is often what you asked for. If you said \"make the cube teal\", the cube is no longer the recorded colour. Graspable does not guess. Under the failed check it asks Intended change? with an Accept new baseline button. - If the change is what you wanted, choose Accept new baseline . The replayed scene becomes what later runs compare against. - If it is not, tell the agent what went ","s":"A different scene is often what you asked for. If you said \"make the cube teal\", the cube is no longer the recorded colour. Graspable does not guess. Under the failed check it asks Intended change? with an Accept new baseline button. - If the change is what you wanted, choose Accept new baseline . The replayed scene becomes what later runs compare against. - If it is not, tell the agent what went wrong, or undo the run.","u":"/docs/xr-tests#accept-a-new-baseline"},{"k":"docs","t":"XR tests","h":"Replay by hand","x":"Choose replay fixture… in the bar above the preview and pick a recording. Afterwards the bar shows a scene verdict and an image verdict. The image verdict can differ even when nothing is wrong, for example when something in the scene animates over time. The completion check uses the scene, not the image.","u":"/docs/xr-tests#replay-by-hand"},{"k":"docs","t":"XR tests","h":"When the recording is empty","x":"A check may say: This means the file has no movement in it. It usually happens when the app crashed in XR while it was being recorded. An empty recording is flagged and never repaired, because no code change can fix it. To fix it, record it again, or delete the file from tests/xr/ .","u":"/docs/xr-tests#when-the-recording-is-empty"},{"k":"docs","t":"XR tests","h":"Test on Quest","x":"For tests on a physical Meta Quest, open your web project and choose Headset above the preview. Use Share on Wi-Fi , or connect by USB and choose Open next to the authorized headset. The headset guide covers connection requirements. In the headset's browser, choose your app's Enter AR or Enter VR button and use the experience while wearing the headset. Check the interactions your app uses: selecti","s":"For tests on a physical Meta Quest, open your web project and choose Headset above the preview. Use Share on Wi-Fi , or connect by USB and choose Open next to the authorized headset. The headset guide covers connection requirements. In the headset's browser, choose your app's Enter AR or Enter VR button and use the experience while wearing the headset. Check the interactions your app uses: selecting and grabbing with controllers or hands, reading nearby text, moving around placed objects, hearing sound and leaving the experience. Watch for pauses or missing objects while moving your head. Repeat after a change that affects those behaviors. Recorded XR tests replay in the desktop emulator. A passing replay does not establish Quest frame rate, hand tracking, passthrough placement or comfort. Keep the replay to catch repeatable regressions, and check the affected behavior on the headset. Unity and Unreal projects use their own native build and device workflow.","u":"/docs/xr-tests#test-on-quest"},{"k":"docs","t":"XR tests","h":"Good to know","x":"- Browser checks serve regular files from the built app. Linked files and folders are refused; copy assets into the project before building. Large files stream as the browser reads them, and stopping the check closes unfinished transfers. - The agent is not allowed to edit files in tests/xr/ by hand. That would let it change the test instead of fixing the app. - The xr check runs in the browser li","s":"- Browser checks serve regular files from the built app. Linked files and folders are refused; copy assets into the project before building. Large files stream as the browser reads them, and stopping the check closes unfinished transfers. - The agent is not allowed to edit files in tests/xr/ by hand. That would let it change the test instead of fixing the app. - The xr check runs in the browser listed in Settings → Components. While that browser is still being downloaded, the check waits for it. - Replays use the emulated headset, not a real one.","u":"/docs/xr-tests#good-to-know"},{"k":"docs","t":"Design panels and controls","x":"Pick a panel, button, slider or label in the preview and change it with the Design panel. Every change is written into your project's code. Design is an editor for the 2D interfaces inside your 3D scene: panels and the buttons, sliders, toggles, segmented choices, progress bars, labels, dividers, spacers, rows and columns on them. Each interface is shown flat and straight on, like a screen in an i","s":"Pick a panel, button, slider or label in the preview and change it with the Design panel. Every change is written into your project's code. Design is an editor for the 2D interfaces inside your 3D scene: panels and the buttons, sliders, toggles, segmented choices, progress bars, labels, dividers, spacers, rows and columns on them. Each interface is shown flat and straight on, like a screen in an interface builder. Click a part, change it in the inspector or ask the agent to change it, and Graspable writes the change into the code that places it. When you go back to the 3D scene or put on a headset, you see the same interface. Design works with the UI kit that comes with the Hands & Spatial UI template. Other web projects, such as Tabletop AR for phones, React XR or a product viewer, can get the same kit: open Design , choose Add the UI kit , and Graspable puts it in a folder of its own, ui/ , without touching your files. The agent then adds a first panel to the scene. The kit comes with the Builder plan and higher, like the template.","u":"/docs/design"},{"k":"docs","t":"Design panels and controls","h":"Open it","x":"Open a web project and choose Design in the header. The window changes for designing: - Left: the scene's interfaces, one per panel, and the parts of the one you are looking at. - Middle: the interface, flat and straight on. In the scene shows it where it stands in 3D instead. - Right of it: the inspector for the selected part. - Far right: the agent, as always. Back to files , or Design again, cl","s":"Open a web project and choose Design in the header. The window changes for designing: - Left: the scene's interfaces, one per panel, and the parts of the one you are looking at. - Middle: the interface, flat and straight on. In the scene shows it where it stands in 3D instead. - Right of it: the inspector for the selected part. - Far right: the agent, as always. Back to files , or Design again, closes it and brings back the files and the view you had before. While Design is open, a click in the preview selects a part instead of pressing it.","u":"/docs/design#open-it"},{"k":"docs","t":"Design panels and controls","h":"Check it for a device","x":"Under the list of interfaces, Checks tells you whether the interface shown can be read and used where it stands. Choose the device above the list: - Headset : whether a finger can reach it (0.35 to 0.7 m away; further away people point and pinch), whether its smallest letters are large enough for their distance, whether its smallest control is big enough for a fingertip (3 cm, better 4 cm) or for ","s":"Under the list of interfaces, Checks tells you whether the interface shown can be read and used where it stands. Choose the device above the list: - Headset : whether a finger can reach it (0.35 to 0.7 m away; further away people point and pinch), whether its smallest letters are large enough for their distance, whether its smallest control is big enough for a fingertip (3 cm, better 4 cm) or for pointing from where it is, whether it fits in view without turning the head, and whether it sits between 10° above and 40° below eye level. - Phone AR : the smallest letters and controls in points on a phone screen held upright, against 11 points for text and 44 for a tap. - Display glasses : glasses such as Meta Ray-Ban Display show a 600 × 600 app, not panels placed in a room. Make an app for them from the Display glasses template. Distances are measured from where a person stands when the immersive session starts, eyes 1.6 m above the floor. Click a check to select the part it is about. Ask the agent to fix it starts a request with the numbers in it, such as making the smallest text 1.4 cm tall. With more than one problem, Ask the agent to fix all sends them in one request.","u":"/docs/design#check-it-for-a-device"},{"k":"docs","t":"Design panels and controls","h":"Select a part","x":"- Click a part in the flat view or in the scene. It gets an outline, and the inspector shows its settings. - Or click it in the list of parts on the left. - Code opens the line that places the part.","u":"/docs/design#select-a-part"},{"k":"docs","t":"Design panels and controls","h":"Work with the agent","x":"The selected part goes with your next message to the agent, so \"make this bigger\" or \"add a volume slider here\" is about that part. The agent's changes show in the flat view as it makes them. Your own edits wait while the agent is working.","u":"/docs/design#work-with-the-agent"},{"k":"docs","t":"Design panels and controls","h":"Change it","x":"The inspector shows the settings each kind of part has: a button's label, size and colours; a slider's range and step; a label's letter size, weight and alignment; a panel's title, position and padding. A setting you have not written shows the kit's default in grey. - Type a value and press Return, or click away. - Drag left and right on the small mark before a number to change it in steps. The ar","s":"The inspector shows the settings each kind of part has: a button's label, size and colours; a slider's range and step; a label's letter size, weight and alignment; a panel's title, position and padding. A setting you have not written shows the kit's default in grey. - Type a value and press Return, or click away. - Drag left and right on the small mark before a number to change it in steps. The arrow keys step it too, ten steps at a time with Shift. - In the flat view, drag the small handles on the selected part's right edge, bottom edge or corner to change its width and height. Parts are centred, so they grow on both sides. The preview follows while you drag, and the view keeps still until you let go; the whole drag is one step to undo. - Select a row, column or panel to see the space between its parts in pink. Drag a pink space to change the gap, or, on a panel, the space at its left, right or bottom edge to change its padding. The new value shows while you drag. - × removes a setting from the code, so the kit's default applies. - A setting that your code computes, such as a slider's value={size} , is shown but not edited here. Change it in the code.","u":"/docs/design#change-it"},{"k":"docs","t":"Design panels and controls","h":"Theme","x":"With nothing selected, the inspector shows the Theme : the colours, spacing and motion that every part of the kit falls back to, from lib/theme.js . Change a colour there, such as accent , and every slider, toggle and button that uses it changes with it. The note after each value in lib/theme.js shows when you point at it. A colour setting of a part has a ◆ list beside it: choose a theme colour to","s":"With nothing selected, the inspector shows the Theme : the colours, spacing and motion that every part of the kit falls back to, from lib/theme.js . Change a colour there, such as accent , and every slider, toggle and button that uses it changes with it. The note after each value in lib/theme.js shows when you point at it. A colour setting of a part has a ◆ list beside it: choose a theme colour to have the part follow the theme (the code says color={theme.accent} ), or Own colour to give it a colour of its own.","u":"/docs/design#theme"},{"k":"docs","t":"Design panels and controls","h":"Motion","x":"Buttons, toggles and sliders move with a spring when they are pressed, pointed at or switched, and a panel comes in with a short movement when it first appears. The inspector's Motion section sets both: - Spring : Quick (the default for controls), Gentle (the default for panels), Bouncy , Slow , or None to move without easing. The curve under the choices shows how the spring settles: Bouncy goes p","s":"Buttons, toggles and sliders move with a spring when they are pressed, pointed at or switched, and a panel comes in with a short movement when it first appears. The inspector's Motion section sets both: - Spring : Quick (the default for controls), Gentle (the default for panels), Bouncy , Slow , or None to move without easing. The curve under the choices shows how the spring settles: Bouncy goes past its end and comes back. - Comes in (for a panel): Rise , Pop , Tilt or None . A panel that your code hides and shows again comes in again, so the pages of a menu animate by themselves. - Play shows the selected part's motion again in the preview. It also plays by itself after you change it. - Animate on a timeline opens Animate with the part selected, to move, turn, scale or fade it with keyframes, for example a panel that slides in during a film. The defaults for the whole app are motion and appear in lib/theme.js . Ask the agent for a spring of your own, such as \"make the buttons a little bouncier\", and it writes { duration, bounce } instead of a name. Projects made before version 0.20.5 have an older kit without springs; the Motion section is not shown for them.","u":"/docs/design#motion"},{"k":"docs","t":"Design panels and controls","h":"Add, move and remove parts","x":"- Add inside (for a panel, row or column) puts a new part at its end. Graspable adds the import if the file needs one. The parts: - Button , Slider and Toggle . - Segmented : one choice out of a few, side by side, such as modes or tabs. Type its options separated by commas. - Progress : a bar that fills from 0 to 1, with an optional label and percentage. - Label : one line of text. - Divider : a t","s":"- Add inside (for a panel, row or column) puts a new part at its end. Graspable adds the import if the file needs one. The parts: - Button , Slider and Toggle . - Segmented : one choice out of a few, side by side, such as modes or tabs. Type its options separated by commas. - Progress : a bar that fills from 0 to 1, with an optional label and percentage. - Label : one line of text. - Divider : a thin line between groups. - Spacer : empty room. - Row and Column : lay parts out across or down. - Add after puts the new part next to the selected one. - Panel beside (for a panel) adds a second interface next to it, starting with a label. - Drag a part in the list of parts to move it: onto the top or bottom edge of another part to put it before or after it, onto the middle of a row, column or panel to put it inside. - Up and Down swap the part with its neighbour. Duplicate copies it, and Delete removes it. - Parts that your code makes in a loop, such as one button per colour, are edited in the code.","u":"/docs/design#add-move-and-remove-parts"},{"k":"docs","t":"Design panels and controls","h":"Keys","x":"While Design is open and you are not typing in a field (the keyboard button at the top of the inspector lists them too): Key What it does --- --- ⌘Z / Ctrl+Z Undo the last design edit ⇧⌘Z / Ctrl+Y Redo it Delete or Backspace Remove the selected part ⌘D / Ctrl+D Duplicate it ⌥↑ / ⌥↓ (Alt on Windows and Linux) Move it before or after its neighbour Arrow keys Move a selected panel by 1 cm, 5 cm with ","s":"While Design is open and you are not typing in a field (the keyboard button at the top of the inspector lists them too): Key What it does --- --- ⌘Z / Ctrl+Z Undo the last design edit ⇧⌘Z / Ctrl+Y Redo it Delete or Backspace Remove the selected part ⌘D / Ctrl+D Duplicate it ⌥↑ / ⌥↓ (Alt on Windows and Linux) Move it before or after its neighbour Arrow keys Move a selected panel by 1 cm, 5 cm with Shift Esc Clear the selection","u":"/docs/design#keys"},{"k":"docs","t":"Design panels and controls","h":"Use a panel on another platform","x":"A selected panel can be written for other platforms, so one design serves several devices. Under Use on another platform in the inspector: - SwiftUI for Apple Vision Pro: a SwiftUI view with stacks, sliders, toggles and buttons, sized so 1 point is 1 mm, on a glass background. Show it in a window or as an attachment in a RealityView. - Unity UI Toolkit for Quest, Android XR and other headsets buil","s":"A selected panel can be written for other platforms, so one design serves several devices. Under Use on another platform in the inspector: - SwiftUI for Apple Vision Pro: a SwiftUI view with stacks, sliders, toggles and buttons, sized so 1 point is 1 mm, on a glass background. Show it in a window or as an attachment in a RealityView. - Unity UI Toolkit for Quest, Android XR and other headsets built with Unity: a UXML file and a USS style sheet. Use them with a UIDocument whose Panel Settings are set to World Space (Unity 6.2 or later), with a scale of 0.001 so 1 pixel is 1 mm. - Display glasses for Meta Ray-Ban Display: a screen of a Display glasses app, built from Meta's Wearables UI toolkit. Glasses have no sliders or switches, so a slider steps with Less and More buttons, a toggle becomes a button that says its state, and a row of buttons goes to the rail at the bottom. Choose one to see the code and Copy it into a project of that kind. Values that your code computes, such as a slider bound to state, become state of the same name, and what a button does is copied as a comment for you or the agent to connect.","u":"/docs/design#use-a-panel-on-another-platform"},{"k":"docs","t":"Design panels and controls","h":"Undo","x":"Undo and Redo at the top of the inspector, or the keys above, step through your design edits. Every edit is also a step in the project's History , so you can go back further from there.","u":"/docs/design#undo"},{"k":"docs","t":"Animation","x":"Animation editing in development, what the current editor can do, and what remains before filmmaking is ready. Version 0.17.2 adds an early animation and filmmaking workspace. Open a web project and choose Animate to edit poses, keyframes, shots and sound. The broader filmmaking workspace remains in development; the limits below describe what this release supports.","u":"/docs/animation"},{"k":"docs","t":"Animation","h":"Edit a take on desktop","x":"Open a web 3D project and choose Animate in its header. The preview stays above the timeline. From version 0.19.1, the animation workspace loads when you choose Animate , alongside preview preparation. Ordinary project editing no longer loads those tools up front. If preview preparation fails, the error stays visible and Animate remains available to try again. 1. With Edit scene selected, click an","s":"Open a web 3D project and choose Animate in its header. The preview stays above the timeline. From version 0.19.1, the animation workspace loads when you choose Animate , alongside preview preparation. Ordinary project editing no longer loads those tools up front. If preview preparation fails, the error stays visible and Animate remains available to try again. 1. With Edit scene selected, click an object in the preview. If needed, choose Enable animation to give it an identity that survives reloading the project. Select parent chooses the authored group containing it. Panels, buttons, sliders, toggles and labels from the UI kit are selected as a whole, so a click on a button selects the button, not a shape inside it. From Design , Animate on a timeline opens Animate with the selected part already chosen. 2. The selected object shows Position , Rotation (°) and Scale together. Change an axis, then press Enter or leave the field to add a key at the current frame. Escape cancels an unfinished field edit. Use Add position key , Add rotation key or Add scale key to keep the current value, or Key all transforms to record the complete pose as one undoable edit. 3. Move the playhead to another frame and set the next pose. Choose Play or scrub the playhead to inspect the motion. 4. Choose Save take . The take stays in the project and can be reopened from the Animation file field. Edit scene intercepts primary clicks on the 3D canvas so selecting an object does not trigger its gameplay action. Click empty space to clear selection; hidden objects and editor helpers are excluded from picking. Interact returns control to the experience. The object menu and timeline remain available for selection. For an object with an animation identity, choose Move , Rotate or Scale and drag a handle in the scene. Releasing records position, rotation and scale at the current frame as one pose; Undo reverses the whole gesture. Press Escape during a drag to cancel. Moving the playhead or leaving edit mode also cancels an unfinished gesture. Move and Rotate use world axes; Scale uses the object's local axes. Handles are unavailable during playback or an agent edit, and are hidden while immersed in XR. These desktop handles do not replace the experimental XR controls below. The initial editor supports position, rotation, scale, opacity and visibility. Linear interpolation moves evenly between poses; Smooth ease in/out starts and ends each interval gently; Hold keeps the earlier pose until the next key. Rotation fields use degrees; the take stores normalized quaternions. A motion path shows the selected object's position track. The pose inspector sits beside the timeline. Large takes keep every keyframe available for selection and editing as you move the playhead. Offscreen tracks defer rendering while you zoom; keyboard focus brings their controls into view. Use Previous key and Next key to move through the selected channel. Curves stay hidden until you choose Show curves ; Hide curves returns the space to tracks. Select a diamond or a curve key to move to its frame. Select Position, Rotation or Scale in the inspector to choose which channel the curve displays. Click a diamond to visit its keyframe. Drag it to change its time, or focus it and use the arrow keys; Shift moves ten frames. Delete removes the selected key. Undo and Redo apply to edits in the current editor session. Dragging shows the destination frame before you release. After retiming, the playhead and keyboard focus follow the moved key. The ruler uses regular frame intervals, and channel names stay visible when you scroll a zoomed timeline. Zoom changes smoothly in fine increments and keeps the visible playhead in place. If the playhead is offscreen or behind the track names, zoom centers on the visible timeline. Pinch or hold Ctrl/Cmd while scrolling to zoom around the pointer. Fit take shows the whole timeline again. Zoom does not change keyframes or mark the take as edited; reduced-motion preferences remove the transition. Use In and Out to limit playback to an inclusive frame range, and enable Loop to repeat it. Full take resets the range. These are session playback controls; they do not trim or change your saved animation. Space plays or pauses when the animation workspace has focus. Text fields and buttons keep their usual keyboard behavior. Version 0.19.1 avoids redundant editor updates between animation frames during looping playback, while preserving audio restarts at loop boundaries. Invalid frame values leave the playhead unchanged. If a take changes on disk, the editor keeps your draft and asks you to load the latest version before saving. Saving is paused while the agent is working. Objects created by repeated components need distinct animation identities for each instance.","u":"/docs/animation#edit-a-take-on-desktop"},{"k":"docs","t":"Animation","h":"Customize easing","x":"Open Easing beside Position, Rotation, Scale or the selected facial, bone or opacity channel. Controls start collapsed. Choose a preset or Custom Bézier for draggable handles and numeric values. Arrow keys adjust a focused handle; Shift takes larger steps. Click Easing again or press Escape inside the panel to close it. Opening and closing do not change the animation. Choosing a curve at an unkeye","s":"Open Easing beside Position, Rotation, Scale or the selected facial, bone or opacity channel. Controls start collapsed. Choose a preset or Custom Bézier for draggable handles and numeric values. Arrow keys adjust a focused handle; Shift takes larger steps. Click Easing again or press Escape inside the panel to close it. Opening and closing do not change the animation. Choosing a curve at an unkeyed frame adds a key using the current pose. Save the take to keep edits. Camera moves have their own collapsed Easing control beside camera positioning. Changing a pose preserves that key’s easing. Custom timing is available for transforms, bone rotation, opacity and facial channels. Visibility uses Hold. This is a timing curve between two values, not a spatial path handle or an editable rig constraint. For controller editing, open Easing in the XR panel. Desktop and XR use the same saved timing curve.","u":"/docs/animation#customize-easing"},{"k":"docs","t":"Animation","h":"Speech, blinks and gaze","x":"Select a bound character with facial controls, then open Speech performance . Import a Rhubarb mouth-cue JSON file and map each spoken shape to an existing control on that character. Matching viseme A through viseme H names are selected automatically; check that each one produces the intended mouth shape before using it. Set the amount, start frame and transition length, then choose Check speech k","s":"Select a bound character with facial controls, then open Speech performance . Import a Rhubarb mouth-cue JSON file and map each spoken shape to an existing control on that character. Matching viseme A through viseme H names are selected automatically; check that each one produces the intended mouth shape before using it. Set the amount, start frame and transition length, then choose Check speech keys and Add speech keys . Save the take to keep the result. The imported times come from the recording, rather than its loudness. They become ordinary named facial keys: scrub, change a weight, retime a sound or undo the whole import. Silence returns the mapped weights to zero. Add the same recording as an audio clip at the matching start frame; importing mouth cues does not add or synthesize audio. Existing mapped channels are protected by default. Replace entire mapped mouth channels explicitly replaces those channels, including keys outside the imported speech range. Other expressions and body animation remain unchanged. A character still needs appropriate mouth, teeth and tongue geometry; this does not create anatomy or infer what an arbitrary blendshape does. Imported characters also expose Rig rotations in the Property menu. Choose a uniquely named bone, edit its local X/Y/Z angles in degrees, and set a keyframe. The saved channel uses a normalized quaternion and interpolates along the shortest rotation. Your source rig determines each bone’s local axes. A keyed bone overrides an imported clip after the clip updates; unkeyed bones keep their existing animation. Removing a channel or stopping authoring restores the pose captured before the override. Duplicate bone names are listed and cannot be bound ambiguously. The picker shows up to 512 unique controls; a hierarchy exceeding the 8,192-object audit limit disables bone binding. Agents can use edit motion with a performance recipe to bake the same explicitly mapped cues, independent blink schedules and quaternion gaze keys. A gaze entry with target rotates an authored object; adding bone rotates that exact unique bone within the authored character. Bone names are scoped to that character and stop at separately authored descendants. The result uses the same revision checks, saved history and undo as manual edits. Gaze holds its first and last keyed pose outside the range, like other rotation tracks. Native control rigs with constraints still need their dependent motion baked; changing one exported deform bone does not recreate a Blender, Unity or Unreal rig solver. This is a timed performance importer, not bundled audio analysis or facial capture. Speech, blink and gaze recipes are independent. An agent can omit speech cues and mouth mappings when changing only eye direction or blink timing; this leaves mouth channels untouched. Rig inspection also stops at 128 nested levels. A model beyond that depth cannot expose safe named bone controls; simplify its hierarchy before binding it. An imported clip updating in the preview does not reset bone angles being typed in the inspector.","u":"/docs/animation#speech-blinks-and-gaze"},{"k":"docs","t":"Animation","h":"Camera shots","x":"Frame the scene in the viewport and choose Capture view at frame… . A shot stores that camera position, direction and lens. Set its In and Out frames, rename it, or change its field of view. Shots cannot overlap. Trim the first shot before capturing the next one. Preview film camera follows the shot sequence while you scrub or play; turn it off to frame another view. Camera cuts never move your he","s":"Frame the scene in the viewport and choose Capture view at frame… . A shot stores that camera position, direction and lens. Set its In and Out frames, rename it, or change its field of view. Shots cannot overlap. Trim the first shot before capturing the next one. Preview film camera follows the shot sequence while you scrub or play; turn it off to frame another view. Camera cuts never move your head in XR. For a moving shot, turn off Preview film camera , frame a second view and choose Capture ending view . The camera moves smoothly between the two poses and reaches the ending pose on the shot's final frame. Ending field of view adds a lens change. Remove camera move restores a static shot. These film-camera moves do not control the XR viewer's head. Choose Edit opening camera or Edit ending camera to see the camera from outside. Move camera and Rotate camera give you handles in the scene, with a live viewfinder showing the composition through that camera. Right-drag orbits your editing view; scroll moves closer or farther away. Release a handle to record the endpoint, or press Escape during a drag to cancel. Done positioning returns to ordinary scene editing. Moving your editing view does not change the shot. An ending pose is added only when you complete a camera gesture. With Edit in XR enabled, squeeze the camera with a controller to move and rotate it, then release to record the endpoint. Squeeze the floating viewfinder to reposition the monitor. Choose Enter camera on the XR panel to look around from the shot's position; Return restores your editing viewpoint. This changes your position and horizontal facing while keeping head tracking live. Bringing the monitor close to your face does not automatically move you. These controls have been checked in the headset emulator; physical Quest testing and hand gestures remain outstanding. The viewfinder is a composition preview, not a final-resolution render or a stereoscopic 360 video export.","u":"/docs/animation#camera-shots"},{"k":"docs","t":"Animation","h":"Add sound","x":"In Sound , choose an audio file already in the project and select Add at playhead . In the desktop app, Import audio copies a file into the project first. The clip initially ends at the shorter of the source file or the end of the take. Each clip shows its waveform and position in the take. Set Start and Frames to move or trim it, and Source in (s) to skip the beginning of the recording. Volume ru","s":"In Sound , choose an audio file already in the project and select Add at playhead . In the desktop app, Import audio copies a file into the project first. The clip initially ends at the shorter of the source file or the end of the take. Each clip shows its waveform and position in the take. Set Start and Frames to move or trim it, and Source in (s) to skip the beginning of the recording. Volume runs from 0 to 1; Fade in and Fade out are measured in frames. These edits leave the original recording intact. Mute a clip to exclude it from playback and export. Audio edits use the same save and undo history as animation keys. Playback follows the playhead, including pauses, seeks and loops. When looping a selected range, preview audio stops at Out before the range starts again. Overlapping clips mix together. Missing files, unsupported audio and trims beyond the source recording produce an error; use Retry audio after replacing a file. This build supports source files up to 32 MB and up to 128 MB of decoded sound per take. Long recordings may need trimming before import. For generated narration or sound effects, open Create sound with connected apps . Connect ElevenLabs in Settings → Connected services , add it to this project, and ask the agent to generate and save an audio file. The existing connected-service approvals and charges apply. The audio picker refreshes as files arrive; add the resulting file to the take. This is an agent-assisted workflow through the connected service, not a separate generation service built into the timeline.","u":"/docs/animation#add-sound"},{"k":"docs","t":"Animation","h":"Export a film","x":"Set In and Out , choose the format and size under Export film , then Export video . Graspable saves the take before rendering its camera cuts, motion and unmuted audio. Completed films are saved in the project, with a local player and a download. They can be selected under Publish → Rendered film . Choose MP4 or WebM for a framed film, 360° MP4 for a surrounding panorama, or Stereo MP4 for separat","s":"Set In and Out , choose the format and size under Export film , then Export video . Graspable saves the take before rendering its camera cuts, motion and unmuted audio. Completed films are saved in the project, with a local player and a download. They can be selected under Publish → Rendered film . Choose MP4 or WebM for a framed film, 360° MP4 for a surrounding panorama, or Stereo MP4 for separate left/right views. Supported Macs also encode native Apple spatial and 360° video. See Share immersive films for YouTube, Vision Pro, headset playback and publishing workflows. Exports are limited to two minutes, 4096 × 2048 and 192 MB locally. Cloud delivery has an 80 MiB limit per media file. Missing audio or an unsupported encoder stops export with an error. Leave an immersive XR session before exporting. If an older project asks you to update its preview tools, do that before rendering immersive formats.","u":"/docs/animation#export-a-film"},{"k":"docs","t":"Animation","h":"Facial performance","x":"Select a character in Animate , bind it with Enable animation if needed, then choose a named control under Property → Facial performance . These are the blendshapes already present in the model, such as jawOpen or eyeBlinkLeft . Set an amount from 0 to 1 and key it at the current frame. Facial channels use the same interpolation, curves, retiming, save and undo as object motion. A control applies ","s":"Select a character in Animate , bind it with Enable animation if needed, then choose a named control under Property → Facial performance . These are the blendshapes already present in the model, such as jawOpen or eyeBlinkLeft . Set an amount from 0 to 1 and key it at the current frame. Facial channels use the same interpolation, curves, retiming, save and undo as object motion. A control applies to matching named blendshapes across that character’s meshes, so the face and teeth can move together even when their internal target indices differ. A separately bound child character keeps its own controls. Other blendshapes retain their existing animation; removing a facial track restores its original weights. The inspector lists up to 256 controls per selected object. Saved channels outside this list remain playable. Agents use morph:<exact blendshape name as the property in the existing motion tools. They can read one channel, edit keys or copy facial curves to other bound characters using revision checks and compact receipts. The target characters need matching blendshapes; Graspable does not invent missing facial anatomy or automatically retarget different facial rigs. A rig without modeled eyes, teeth or articulated fingers still needs a better source asset. These controls are part of the connected authoring preview and frame renderer. A published interactive scene must consume the saved take in its own playback code. Motion-only USD exports preserve the named curves as custom attributes and report that they are not bound to a USD skeleton.","u":"/docs/animation#facial-performance"},{"k":"docs","t":"Animation","h":"Editing with agents","x":"Saving uses a revision check and a unique operation ID. Retrying the same save does not apply it twice. Other authenticated agents can read and edit the same take through the motion tools; a stale edit is refused, and your local draft stays available. Current editing is coordinated by one running Graspable agent. Remote invitations, participant presence and simultaneous conflict merging are not ye","s":"Saving uses a revision check and a unique operation ID. Retrying the same save does not apply it twice. Other authenticated agents can read and edit the same take through the motion tools; a stale edit is refused, and your local draft stays available. Current editing is coordinated by one running Graspable agent. Remote invitations, participant presence and simultaneous conflict merging are not yet available. Saved edit history shows recent accepted changes, including object identities, keyframes, shots and sound settings. Starting an agent request saves the open animation draft first. If it cannot be saved, the request stays in the composer so you can resolve the problem. Each new request receives recent saved-edit summaries and recent project versions, including when you continue a conversation. The run's Context pane shows this information. Session labels describe the caller; they are not verified participant identities. This context is bounded: up to 32 recent saved operations, subject to a size limit, and six recent project versions. The current project files remain authoritative. Older changes remain available through project history, but the agent does not automatically receive the entire history or unsaved source-code drafts.","u":"/docs/animation#editing-with-agents"},{"k":"docs","t":"Animation","h":"Repeat motion across objects","x":"Select an animated object and open Copy animation to objects . Choose destination objects with animation identities, the channels to copy, and any frame stagger or position offset. Random variation uses a seed, so the same settings produce the same result. Position values are copied from the source with the chosen offsets; this does not automatically preserve each destination's starting position. ","s":"Select an animated object and open Copy animation to objects . Choose destination objects with animation identities, the channels to copy, and any frame stagger or position offset. Random variation uses a seed, so the same settings produce the same result. Position values are copied from the source with the chosen offsets; this does not automatically preserve each destination's starting position. Preview the counts before applying. Existing destination channels are protected unless you explicitly choose replacement. Applying changes the draft as one undoable action; save to share it with agents. Agents can use the same bulk operation through the motion editing tool. This build expands the result into editable tracks, with a limit of 4,096 tracks and 100,000 keyframes per take. Millions of objects will need an instanced, procedural animation system; that capability is not implemented here.","u":"/docs/animation#repeat-motion-across-objects"},{"k":"docs","t":"Animation","h":"Edit in XR","x":"Enable Edit in XR , then enter VR in the preview. A panel in the scene provides previous/next frame, play/pause, Set key or Update key, undo and save, plus shot navigation, capture and recentering. Choose Move , Rotate or Scale , then use the axis −/+ controls to change the selected object and key that channel. The step control cycles through fine and coarse adjustments. Choose Easing to open pres","s":"Enable Edit in XR , then enter VR in the preview. A panel in the scene provides previous/next frame, play/pause, Set key or Update key, undo and save, plus shot navigation, capture and recentering. Choose Move , Rotate or Scale , then use the axis −/+ controls to change the selected object and key that channel. The step control cycles through fine and coarse adjustments. Choose Easing to open presets; Custom reveals the four handle controls. Choose Apply easing to keep the draft, or Done to close without applying it. While editing a moving camera, these controls apply to that shot. Each new object or camera selection starts with easing hidden. Point at its timeline to seek. Point at an animated object to select it. Hold the controller’s squeeze button to manipulate the selected Move, Rotate or Scale channel, then release to add one undoable key. Scaling changes all three axes together according to controller distance. Playback, a changed frame, an edit elsewhere, loss of tracking or leaving XR cancels an unfinished gesture. Desktop and XR changes use the same keys and saved take. For a real headset on the same Wi-Fi, enable Edit in XR and choose Connect headset . Open the QR code or editing link in the headset browser, accept the local preview certificate if asked, then enter VR. Keep the desktop take open: it remains the authority for edits and saves. The link expires after 30 minutes; Disconnect headset editing , closing the take or stopping its preview removes editing access. Anyone with the link can edit this take while it is active. The real-headset connection has been exercised with a browser through the native WebXR path. Physical Quest 3/3S controller comfort, hand input and readability still need hardware testing. Headset audio monitoring and autonomous editing without the desktop are not implemented. The current XR panel uses controller rays and squeeze gestures; hand manipulation and a fuller immersive workspace remain in development.","u":"/docs/animation#edit-in-xr"},{"k":"docs","t":"Animation","h":"Filmmaking work still ahead","x":"The timeline can export a saved take as USDA animation and cameras or an OpenTimelineIO shot/audio schedule. Save the take before exporting. Agents can also prepare supported OTIO editorial changes for review and apply them through the same revision-checked motion operations. A take export does not reconstruct scene geometry or rigs. For actual scene inspection, layered edits and supported USD/glT","s":"The timeline can export a saved take as USDA animation and cameras or an OpenTimelineIO shot/audio schedule. Save the take before exporting. Agents can also prepare supported OTIO editorial changes for review and apply them through the same revision-checked motion operations. A take export does not reconstruct scene geometry or rigs. For actual scene inspection, layered edits and supported USD/glTF model conversion, use the OpenUSD inspector in Assets. See interchange files for the take formats and limitations. Agents now read a small film overview, then request the channels, keys or shots needed for an edit. They can check a revision or ask which items changed without repeatedly reading the whole take. Successful edits return short receipts; the saved motion document remains editable JSON. Full rig and clip editing in the timeline, spatial path handles, broader collaboration, stereo panoramic capture and longer production export workflows remain unfinished. The controls above are development features, not a claim that the filmmaking milestone is complete. Native Unity and Unreal workflows will use their own editable animation assets and engine tools; they are not implemented by the web timeline. For the current motion document format and read-only agent tools, see motion inspection.","u":"/docs/animation#filmmaking-work-still-ahead"},{"k":"docs","t":"OpenUSD scenes","x":"Inspect USD scenes, save edits in separate layers, and exchange supported models, materials, cameras and rigs with glTF. OpenUSD support is available from 0.17.2 . You can inspect a scene, save changes in a separate layer and convert supported scene data for use in web projects.","u":"/docs/openusd"},{"k":"docs","t":"OpenUSD scenes","h":"Asset library","x":"On Pro and Studio , open a project’s Assets tab and choose Asset library . Search by name or supplied description, filter by source and format, then select a result to see its supplied preview, download size, license and source. The library includes NVIDIA SimReady OpenUSD assets and Scenario’s 3D collection from Hugging Face. Scenario models and splats Scenario supplies GLB models, Blender scenes","s":"On Pro and Studio , open a project’s Assets tab and choose Asset library . Search by name or supplied description, filter by source and format, then select a result to see its supplied preview, download size, license and source. The library includes NVIDIA SimReady OpenUSD assets and Scenario’s 3D collection from Hugging Face. Scenario models and splats Scenario supplies GLB models, Blender scenes, SPZ splats, PLY, voxel and FBX files. When an example has several formats, choose Import format to download only the one you need. Import asset copies it into your project with its license, attribution and source record. GLB files open in the model preview; other formats remain available in Assets for use by your agent or their native tools. They do not require the OpenUSD runtime. For a GLB up to 32 MiB, choose Preview in 3D before importing. Browsing loads only supplied images; one selected model loads when you ask. Switching assets or closing the preview cancels loading and releases its graphics resources. The viewer rests when unchanged, pauses offscreen, and adjusts resolution during slow animation playback. Sharpness recovers as playback becomes smooth again. Recent verified model downloads are held in a bounded cache so importing a previewed file does not download it again. A larger model can still be imported within the 64 MiB file limit. These are AI-generated assets. Supplied previews help you choose, but do not guarantee matching appearance in every renderer, rigging, animation or XR performance. Inspect the result in your scene before publishing. Blender files are copied without executing scripts; open them with auto-execution disabled. Some source scenes may need missing textures. Files above 64 MiB are listed with their source link but cannot be imported through this workflow. The main Scenario collection uses CC BY 4.0 and requires credit to Scenario . Graspable preserves this attribution beside each import. Its separate Hunyuan collection has different licenses and is not included. NVIDIA SimReady packages Import and inspect downloads that asset’s USD package, checks its files and dependencies, and opens the USD inspector. You can cancel before the import finishes. An incomplete download does not leave a partial asset in your project. The original scene, referenced files, textures, license and attribution are kept together. The library contains 2,575 catalog entries from a pinned NVIDIA release. Some packages exceed the current import limits: 256 MiB total, 64 MiB for one file, or 1,021 source files. Their source link stays available for a native workflow. Importing a package does not guarantee that every shader, physics setup or face can be converted for the web. The 3D preview opens automatically. Check any conversion notes and inspect the result before adding it to a scene. A conversion failure leaves the original USD available. These assets are published by NVIDIA under CC BY 4.0 . Pro provides the integrated library workflow; the original assets remain openly available on Hugging Face. Keep the included attribution when sharing an experience and describe any modifications. The library download itself does not spend Credits; asking an agent to work with it uses your selected AI account as usual. Agents using Graspable’s built-in harness, Claude Code or Codex can search the same library and import an asset when your signed-in plan grants access. They receive names, sizes and file paths rather than model data in their conversation. You can then ask them to inspect parts, create an edit layer or build a supported web model.","u":"/docs/openusd#asset-library"},{"k":"docs","t":"OpenUSD scenes","h":"Try the examples","x":"- Clear the aisle: move three pallets into their bays, watch the editable camera rehearsal, or enter VR. Continue from its source. - Object study: inspect an imported extinguisher’s materials and geometry, then view it at life size in VR. Continue from its source. Both retain the NVIDIA source packages and attribution. Their web models are separate optimized conversions.","u":"/docs/openusd#try-the-examples"},{"k":"docs","t":"OpenUSD scenes","h":"Open a scene","x":"In Assets , select a .usd , .usda , .usdc or .usdz file to open its 3D preview in the sidebar. Choose Open scene for a larger view and editing tools. Adding a USD scene copies its local dependencies with it. Keep its referenced layers and textures beside the source when importing. The scene opens with a large 3D preview. Drag to orbit and scroll to zoom. Click an object to select it and open its e","s":"In Assets , select a .usd , .usda , .usdc or .usdz file to open its 3D preview in the sidebar. Choose Open scene for a larger view and editing tools. Adding a USD scene copies its local dependencies with it. Keep its referenced layers and textures beside the source when importing. The scene opens with a large 3D preview. Drag to orbit and scroll to zoom. Click an object to select it and open its editing tools. The same object is highlighted in the preview and in the Objects list. Dragging still orbits the camera. Use Select parent to choose its containing group, Focus selection to bring the selected geometry into view, or Deselect to clear it. You can also open Edit scene and select from the hierarchy. Variants, source layers and scene details remain available without filling the initial view. Use More objects or More properties to continue through a large scene. Repeated copies that the preview draws together are chosen from the Objects list. Graspable includes the OpenUSD runtime in supported desktop builds; no separate Python installation is needed. If it is missing or unavailable for your platform, the inspector explains why. Linux ARM64 currently has no bundled OpenUSD runtime. The rest of Graspable remains available.","u":"/docs/openusd#open-a-scene"},{"k":"docs","t":"OpenUSD scenes","h":"Save changes in a layer","x":"Open Edit scene , choose an object and queue the changes you want: - Position, rotation and scale sets a position, rotation or scale correction after the object's existing transform operations. Only rows you change are written. The original position and pivot setup stay in the source. - Hide and Show change visibility. - Under Advanced properties , edit a supported value and choose Add change . Le","s":"Open Edit scene , choose an object and queue the changes you want: - Position, rotation and scale sets a position, rotation or scale correction after the object's existing transform operations. Only rows you change are written. The original position and pivot setup stay in the source. - Hide and Show change visibility. - Under Advanced properties , edit a supported value and choose Add change . Leave the time field empty for a default value, or enter a USD time code to set an animation sample. Array values use brackets, such as [1, 2, 3] . - Inspect samples lists authored time samples. Select one to edit its value and time, or queue Remove sample . - Under Variations , choose one of the scene's existing alternatives, or None to block its current selection in the new layer. Queued changes have not been saved. Undo last change removes the last one; Discard changes clears the queue. Choose Save changes . To change the destination, open Save as a separate edit layer and enter a new .usda path. The new layer refers to the original scene and supplies stronger values for the properties you changed. The inspector switches to this edited scene. Original layers, textures, rigs and geometry stay intact. Each save is recorded in project history. Binary assets use project History for restoration; run undo covers text edits. If the source or a dependency changed since you opened it, saving is refused until you refresh and review the current scene. A time code uses the scene's time-code rate; it is not necessarily a second or a frame in your film take. Editing a USD sample does not automatically edit film.motion.json .","u":"/docs/openusd#save-changes-in-a-layer"},{"k":"docs","t":"OpenUSD scenes","h":"Preview and export","x":"Previews draw the USD scene itself, in the app, usually in under a second. Graspable reads the scene's layers, payloads and textures from your project and composes them with an OpenUSD viewer that runs in the app (Hugging Face's USD web viewer). Nothing is converted, and previewing does not add files or commits to your project. That viewer shows meshes, the basic shapes, instancing, UsdPreviewSurf","s":"Previews draw the USD scene itself, in the app, usually in under a second. Graspable reads the scene's layers, payloads and textures from your project and composes them with an OpenUSD viewer that runs in the app (Hugging Face's USD web viewer). Nothing is converted, and previewing does not add files or commits to your project. That viewer shows meshes, the basic shapes, instancing, UsdPreviewSurface and supported MDL materials with PNG, JPEG and WebP textures. It draws a still scene. Click a mesh to select that object. When a scene has a time range, Play its animation switches to a converted preview that plays animation, skeletons and blend shapes. The first conversion takes longer; an unchanged scene reuses it. That view also selects objects by click, and Fit model frames the whole model. Not shown in this view lists what the still viewer left out, such as curves, points or materials it cannot read. The converted preview lists Preview differences . A failed preview offers Try again and leaves the source intact. Open Export when you need a file. Save queued changes before using Export scene : Format Use --- --- USD text (.usda) A readable USD file for further editing USD binary (.usdc) A binary USD layer for storage and exchange USD package (.usdz) A package containing the scene and its supported local dependencies Web model (.glb) A single web asset with supported geometry, materials, textures, rigs and animation Conversion writes a separate file. Native USD conversion preserves composition by default. Enable Flatten USD composition on export when you need resolved scene data in one layer; flattening removes the original composition structure from that derived output. To bring a web model into USD, select a GLB or glTF asset and choose Export as USD . Its supported meshes, hierarchy, materials, camera and rig data become USD scene data. Animation clips become named variants. A USD-to-GLB conversion uses the currently selected variant.","u":"/docs/openusd#preview-and-export"},{"k":"docs","t":"OpenUSD scenes","h":"Save part of a scene as an asset","x":"Select an object in the preview or the Objects list, then choose Save as asset… . The export panel includes Selected object and its children . Choose GLB, USDA, USDC or USDZ, give the file a name under Save as , and choose Save asset . Save or discard queued scene changes first. The original scene stays open after the new asset is saved. Keep the source asset’s license and attribution when sharing","s":"Select an object in the preview or the Objects list, then choose Save as asset… . The export panel includes Selected object and its children . Choose GLB, USDA, USDC or USDZ, give the file a name under Save as , and choose Save asset . Save or discard queued scene changes first. The original scene stays open after the new asset is saved. Keep the source asset’s license and attribution when sharing an extracted model. The new file keeps the selected object's placement and animation, including parent transforms, required materials, textures and rig data. It retains the original USD object paths. A click on an object generated by a PointInstancer selects the whole instancer; individual particles cannot be separated in this view. A face subset must be exported with its containing mesh. Selection exports resolve the current layers and variations into one scene. They do not retain unused variations or the original layer structure. If a selected object depends on another scene object that cannot be safely included, Graspable asks you to select a parent containing both. Material-only selections can be saved as USD; GLB needs geometry. GLB exports follow the conversion limits below.","u":"/docs/openusd#save-part-of-a-scene-as-an-asset"},{"k":"docs","t":"OpenUSD scenes","h":"What survives conversion","x":"The converter supports triangle meshes, bounded concave-polygon triangulation (with reported approximations for non-planar faces), hierarchy, local transforms, normals, the first UV set, display colors and opacity, material subsets, perspective cameras and basic punctual lights. Cube, sphere, capsule, cylinder and cone primitives can be tessellated. Native instances and supported PointInstancer sc","s":"The converter supports triangle meshes, bounded concave-polygon triangulation (with reported approximations for non-planar faces), hierarchy, local transforms, normals, the first UV set, display colors and opacity, material subsets, perspective cameras and basic punctual lights. Cube, sphere, capsule, cylinder and cone primitives can be tessellated. Native instances and supported PointInstancer scenes can be expanded into web meshes, including bounded instance animation. Materials use USD Preview Surface. Supported PNG/JPEG texture maps include base color, emission, normal, occlusion, alpha and shared metallic/roughness packing. Specialized shader graphs and unusual texture packing still need review. Display colors provide vertex colors when a mesh has no explicit material; an existing bound material keeps precedence. Rigged characters use actual USD skeletons, joint bindings and animation, with up to four joint influences per vertex. Dense and sparse morph targets become blend shapes, including animated weights. Morph-only models receive an identity skeleton so USD's deformation evaluator can animate them. Translation, rotation and scale animation can be exchanged. Cubic glTF animation is baked at 60 Hz; hold transitions use a reported approximation of no more than one microsecond. This does not retarget a character or add a timeline editor for individual bones. Subdivision surfaces are refused when they cannot be evaluated: exporting their coarse control mesh would change the shape. Corrective morph inbetweens, arbitrary shaders, animated topology or visibility, vertex-cache animation, animated camera lenses, nested or skinned PointInstancer prototypes and per-instance primitive variables have conversion limits. Original USD assets retain their native data. Read the report and review the converted scene in its destination renderer before using it in a finished film. Web models keep camera lenses and light settings from the scene's start time. The report flags changing camera apertures, clipping ranges, projection, light color, brightness and spot cones, as well as visibility changes. These properties are not animated in the converted GLB.","u":"/docs/openusd#what-survives-conversion"},{"k":"docs","t":"OpenUSD scenes","h":"Work with agents","x":"Agents can read the stage overview, then request one object, property, time-sample page, layer list or variant set. A revision covers the source and its local dependencies. Repeating a read with that revision can return only “unchanged”; a change query lists affected object paths so the agent can read those objects again. Edits use the same revision check and separate output layers as the inspecto","s":"Agents can read the stage overview, then request one object, property, time-sample page, layer list or variant set. A revision covers the source and its local dependencies. Repeating a read with that revision can return only “unchanged”; a change query lists affected object paths so the agent can read those objects again. Edits use the same revision check and separate output layers as the inspector. Agents can set typed attributes or samples, remove a sample, choose an existing variant, change visibility, add local transform corrections and extract a selected object into a separate asset. They receive output paths, hashes, counts and bounded conversion reports instead of a full scene copied into the conversation. Large reports state how many entries were omitted; mappings can be inspected on individual USD objects. See the USD API and MCP tools. Current limits include 64 MiB per project file, 256 MiB across a dependency graph, 2,048 dependency files and 128 edits in one save. Web conversion has additional geometry, hierarchy and animation-sample limits; exceeding them produces an error. Remote dependencies, unsafe paths and symlink resources are refused. For film-take animation and editorial exports, see Animation. A take's USDA export contains its motion and cameras; use the Assets workflow here when you need to convert actual scene geometry and rigs.","u":"/docs/openusd#work-with-agents"},{"k":"docs","t":"Share immersive films","x":"Export 360 video, stereo films and Apple spatial video, or publish a film player on graspable.cloud. Film delivery is available from 0.17.2 . A rendered film can be watched on a screen or in a supported headset. Publish the interactive project instead when viewers need to move through the scene or interact with its objects.","u":"/docs/film-delivery"},{"k":"docs","t":"Share immersive films","h":"Render and save","x":"Open Animate , choose your take and set In and Out . Under Export film , choose a format and size, then Export video . Graspable saves the take first, renders its camera sequence and mixes the unmuted sound clips. Keep the preview open and leave any active XR session before rendering. Format What it contains Where to use it --- --- --- MP4 A framed 2D film with sound Websites and ordinary video pl","s":"Open Animate , choose your take and set In and Out . Under Export film , choose a format and size, then Export video . Graspable saves the take first, renders its camera sequence and mixes the unmuted sound clips. Keep the preview open and leave any active XR session before rendering. Format What it contains Where to use it --- --- --- MP4 A framed 2D film with sound Websites and ordinary video platforms 360° MP4 A monoscopic, full-sphere equirectangular film with spherical metadata YouTube 360 and compatible headset players Stereo MP4 Full side-by-side left and right camera views Graspable's headset player and compatible stereo players Apple spatial video A framed stereoscopic MV-HEVC MOV with spatial metadata Apple Vision Pro and compatible Apple playback Apple 360° video Monoscopic HEVC with Apple's projected-media metadata Files and Quick Look on visionOS 26 or later WebM A framed 2D film with sound Compatible web browsers Stereo rendering uses two distinct cameras. Eye separation controls their distance in millimeters; it does not add depth to an existing flat video. Apple spatial export needs a fixed lens across the selected range. Split a film with lens changes into separate exports. Apple spatial encoding requires a supported Mac running macOS 15 or later; Apple 360 encoding requires macOS 26 or later. Graspable checks the encoder before you start. The encoder is included in supported desktop builds. Windows and Linux can render the browser formats; native Apple encoding is unavailable there. The saved film appears in the project's files and in Publish → Rendered film . Download video saves a local copy. Preview film player opens its player. If native Apple encoding is canceled or fails, the completed browser video remains saved and available.","u":"/docs/film-delivery#render-and-save"},{"k":"docs","t":"Share immersive films","h":"Publish on graspable.cloud","x":"Choose Publish , select Rendered film , then choose the saved film. Set the address and visibility and publish. This creates a player with playback, seeking, sound, full screen and downloads. It uses the same publishing account and permissions as an interactive project. Publishing a film to an existing project address replaces the content served at that address. Viewers can drag to look around a 3","s":"Choose Publish , select Rendered film , then choose the saved film. Set the address and visibility and publish. This creates a player with playback, seeking, sound, full screen and downloads. It uses the same publishing account and permissions as an interactive project. Publishing a film to an existing project address replaces the content served at that address. Viewers can drag to look around a 360° film on a screen. In a compatible WebXR browser, Watch in headset starts immersive viewing. Stereo sources supply a different image to each eye. A native Apple master, when attached, is offered as a separate download. Browser support varies; the downloadable file remains available when immersive playback is unavailable. On an iPhone or another supported phone, open the published HTTPS address and choose Enable motion . Allow motion access when Safari asks, then turn or tilt the phone to look around a 180° or 360° film. Center view makes your current phone position face the film's starting direction. Portrait and landscape both work; Disable motion keeps your current view and returns to dragging. If permission is denied or the phone supplies no sensor data, you can still drag. Motion viewing does not require WebXR and does not control the view during a headset session. A spherical video lets viewers turn their heads, but it cannot reveal new views when they walk sideways. For that freedom of movement, publish the original interactive scene. Generated render folders are omitted from a film's downloadable source bundle; the scene, takes and original assets remain available for re-rendering.","u":"/docs/film-delivery#publish-on-graspable-cloud"},{"k":"docs","t":"Share immersive films","h":"Upload to YouTube 360","x":"1. Export 360° MP4 . Choose 4096 × 2048 when the scene and device can handle it. 2. Download the MP4 and upload it through YouTube Studio. Graspable adds spherical projection metadata during export. 3. Wait for processing, then confirm that the player offers panning and shows the correct orientation. Check it in a headset before publishing widely. YouTube says 360° processing can take up to an hou","s":"1. Export 360° MP4 . Choose 4096 × 2048 when the scene and device can handle it. 2. Download the MP4 and upload it through YouTube Studio. Graspable adds spherical projection metadata during export. 3. Wait for processing, then confirm that the player offers panning and shows the correct orientation. Check it in a headset before publishing widely. YouTube says 360° processing can take up to an hour. Its 360° upload guide describes playback verification. Graspable's metadata has been checked locally; automated YouTube ingestion testing is not included.","u":"/docs/film-delivery#upload-to-youtube-360"},{"k":"docs","t":"Share immersive films","h":"Watch on Apple Vision Pro","x":"Export Apple spatial video for a framed film with stereoscopic depth, or Apple 360° video for a surrounding panorama. Download the MOV and transfer it using AirDrop or iCloud Drive. Open it in Files or Quick Look on a compatible Vision Pro. Apple's projected-media guidance covers panoramic playback on visionOS 26 and later. These are different deliveries: a framed spatial film is not a 360° film. ","s":"Export Apple spatial video for a framed film with stereoscopic depth, or Apple 360° video for a surrounding panorama. Download the MOV and transfer it using AirDrop or iCloud Drive. Open it in Files or Quick Look on a compatible Vision Pro. Apple's projected-media guidance covers panoramic playback on visionOS 26 and later. These are different deliveries: a framed spatial film is not a 360° film. This build does not author Apple Immersive Video packages or stereo omnidirectional panoramas. Existing correctly formatted 180° and stereo spherical sources can be described in a film manifest for the web player; the timeline renderer currently generates mono 360° only.","u":"/docs/film-delivery#watch-on-apple-vision-pro"},{"k":"docs","t":"Share immersive films","h":"Work with agents","x":"Agents use inspect film to list a few deliveries, inspect a manifest or obtain the exact revision of a source video. encode spatial video converts a supported source with a revision check and returns a compact receipt. Video bytes and generated player code do not enter these tool responses. The local API can also attach a native master to a delivery. Editing an existing film still uses ordinary gu","s":"Agents use inspect film to list a few deliveries, inspect a manifest or obtain the exact revision of a source video. encode spatial video converts a supported source with a revision check and returns a compact receipt. Video bytes and generated player code do not enter these tool responses. The local API can also attach a native master to a delivery. Editing an existing film still uses ordinary guarded file edits and project history. A delivery manifest is a small .film.json file: Paths refer to files inside the project. Optional poster and download fields attach a poster or native master. projection also supports flat and equirectangular-180 ; stereo also supports left-right and top-bottom . Imported videos must already match the declared layout. Changing a manifest does not convert their pixels or add platform metadata.","u":"/docs/film-delivery#work-with-agents"},{"k":"docs","t":"Share immersive films","h":"Current limits","x":"Timeline rendering is limited to two minutes, 4096 × 2048 and 192 MB per local export. Cloud film delivery currently limits each media file to 80 MiB; shorten the range or reduce the size if needed. Browser codec availability can also limit export. Capture renders the 3D scene; HTML overlays and arbitrary post-processing pipelines are not reconstructed. Keyed motion uses exact frames, while live p","s":"Timeline rendering is limited to two minutes, 4096 × 2048 and 192 MB per local export. Cloud film delivery currently limits each media file to 80 MiB; shorten the range or reduce the size if needed. Browser codec availability can also limit export. Capture renders the 3D scene; HTML overlays and arbitrary post-processing pipelines are not reconstructed. Keyed motion uses exact frames, while live physics and procedural effects may vary. A failed or canceled render does not publish anything. Desktop rendering, file metadata, native encoding and emulated headset playback have automated and browser checks. Physical Quest and Vision Pro playback still require hardware validation. OpenUSD preserves editable scene data; it is not a video codec. See OpenUSD scenes for interchange support and conversion limits.","u":"/docs/film-delivery#current-limits"},{"k":"docs","t":"Fast mode","x":"Quicker edits, code search by meaning and shorter build output for the agent. Comes with the Pro plan and uses Credits. Fast mode makes the agent quicker on larger projects. It comes with the Pro plan and higher, and uses Credits. On a new or small project it changes little: there the time goes to the AI model itself and to checking the app, and neither is what Fast mode speeds up. It pays off whe","s":"Quicker edits, code search by meaning and shorter build output for the agent. Comes with the Pro plan and uses Credits. Fast mode makes the agent quicker on larger projects. It comes with the Pro plan and higher, and uses Credits. On a new or small project it changes little: there the time goes to the AI model itself and to checking the app, and neither is what Fast mode speeds up. It pays off when files are long and the code is too large to read through.","u":"/docs/fast-mode"},{"k":"docs","t":"Fast mode","h":"Turn it on","x":"1. Sign in under Settings → Graspable account . 2. Open Settings → Fast mode and tick Use Fast mode . It applies to every run you start after that, on this computer and in Cloud Agents.","u":"/docs/fast-mode#turn-it-on"},{"k":"docs","t":"Fast mode","h":"What it does","x":"Step What changes Shown in the run as --- --- --- Fast edit The agent writes only the lines that change. The edit is merged into the file in about a second, instead of the agent rewriting the file fast edit Code search The agent asks for code by what it does (\"where is the tape length computed\") and gets the right files and lines back fast search Shorter output Long command output, such as a build","s":"Step What changes Shown in the run as --- --- --- Fast edit The agent writes only the lines that change. The edit is merged into the file in about a second, instead of the agent rewriting the file fast edit Code search The agent asks for code by what it does (\"where is the tape length computed\") and gets the right files and lines back fast search Shorter output Long command output, such as a build log, is cut down to the lines that matter before the agent reads it fast shortened output Handover After a project moves to the cloud or back, the earlier conversation is cut down to what matters for your next request fast shortened output Progress check When a fix has failed twice, Graspable checks whether the agent is repeating itself, and stops the run if it is reflex Each step in the run shows the Credits it used.","u":"/docs/fast-mode#what-it-does"},{"k":"docs","t":"Fast mode","h":"What it costs","x":"An edit, a shortened output or a progress check costs 0.1 to 0.3 Credits on a normal file, and more on a very large one. A code search costs about 1 Credit. A step that fails costs nothing.","u":"/docs/fast-mode#what-it-costs"},{"k":"docs","t":"Fast mode","h":"Which agents it works with","x":"Agent Fast edit Code search Shorter output Progress check --- --- --- --- --- Graspable agent Yes Yes Yes Yes Claude Code No. It uses its own editing Yes No Yes Codex CLI No No No Yes","u":"/docs/fast-mode#which-agents-it-works-with"},{"k":"docs","t":"Fast mode","h":"When it is not available","x":"If you run out of Credits, or your plan does not include Fast mode, the run carries on without it. The agent uses its ordinary edit and search tools. Nothing fails.","u":"/docs/fast-mode#when-it-is-not-available"},{"k":"docs","t":"Fast mode","h":"Privacy","x":"For a Fast mode step, the file being edited, the search question with the lines the search reads, or the command output is sent to Graspable and to our provider for that step. It is not stored by Graspable. Without Fast mode none of this leaves your computer except to your own AI account.","u":"/docs/fast-mode#privacy"},{"k":"docs","t":"Agent spending","x":"Let the agent pay for outside services it needs, from a wallet of its own, inside limits you set. Experimental. Some requests need something that costs money: a web search, a dataset, a model, compute. With agent spending on, the agent can find services that are paid per call and buy from them, inside limits you set. It is experimental and off until you switch it on. Three things to know first: - ","s":"Let the agent pay for outside services it needs, from a wallet of its own, inside limits you set. Experimental. Some requests need something that costs money: a web search, a dataset, a model, compute. With agent spending on, the agent can find services that are paid per call and buy from them, inside limits you set. It is experimental and off until you switch it on. Three things to know first: - The agent pays the service directly , from a wallet of its own. Graspable takes nothing from these payments, is not told what was bought, and keeps no record of it. - The limits are enforced by Graspable, not by the AI model , before anything is signed. - Everything is written down for you , on your computer only: what was bought or refused, for which request, and how it was decided.","u":"/docs/agent-spending"},{"k":"docs","t":"Agent spending","h":"Switch it on","x":"Open Settings → Agent spending . 1. Make the hot wallet. Graspable makes a new wallet for the agent on this computer and keeps its key in your keychain. It is yours, and you can show its key at any time. It is called hot because the key is on this computer and not in a hardware wallet, which is what lets the agent pay without you signing each purchase; keep in it only what you would let the agent ","s":"Open Settings → Agent spending . 1. Make the hot wallet. Graspable makes a new wallet for the agent on this computer and keeps its key in your keychain. It is yours, and you can show its key at any time. It is called hot because the key is on this computer and not in a hardware wallet, which is what lets the agent pay without you signing each purchase; keep in it only what you would let the agent spend. It is separate from your personal wallets, and Graspable never asks for those. 2. Send it USDC on the Base network ( Copy address ). Other coins and other networks will not arrive. The wallet can spend only what you send it, so keep in it what you would keep in an app. 3. Set the limits : for one purchase, one request, and one day. 4. Choose how readily it pays , and tick Let the agent pay for outside services . The wallet stays on your computer. Agent spending does not work in Cloud Agents.","u":"/docs/agent-spending#switch-it-on"},{"k":"docs","t":"Agent spending","h":"What the agent can do","x":"It gets two tools. One searches a public catalogue of services that are paid per call and shows each one's price and how many different payers used it in the last 30 days. The other calls a service: the agent says what it needs and why, and Graspable decides. The agent never holds the wallet's key and never builds a payment itself.","u":"/docs/agent-spending#what-the-agent-can-do"},{"k":"docs","t":"Agent spending","h":"How a purchase is decided","x":"In this order: 1. Limits. A request or a day that has used its budget buys nothing more. A service that asks more than its listed price is refused. Only exact amounts of USDC on Base are paid, only to services from the catalogue or ones already paid before. 2. Who is paid. Separate from how much. See below. 3. The judge. System One answers four questions: does your request need this, would a free ","s":"In this order: 1. Limits. A request or a day that has used its budget buys nothing more. A service that asks more than its listed price is refused. Only exact amounts of USDC on Base are paid, only to services from the catalogue or ones already paid before. 2. Who is paid. Separate from how much. See below. 3. The judge. System One answers four questions: does your request need this, would a free tool the agent already has do, is the price in line with the alternatives, and does the reason come from your request or only from text the agent read on the way. It can make the agent stricter than the limits, never looser. 4. You , whenever the setting, the service or the judge says so. How readily it pays What happens --- --- Ask me every time Nothing is paid without your click Save money Pays only when the judge is very sure the request needs it and no free way exists. Asks above a small amount and for a service it has not paid before Balanced Pays for what the request needs within the budget. Asks for a service it has not paid before Finish fast Pays without asking inside the budget. Still refuses what the request does not need You are always asked when: - the amount is above your limit for one purchase; - the reason to buy seems to come from text the agent read, not from your request; - fewer than three others have paid the service in the last 30 days; - no judge with measured confidence is available; - your approval setting is Strict.","u":"/docs/agent-spending#how-a-purchase-is-decided"},{"k":"docs","t":"Agent spending","h":"Who the agent may pay","x":"A budget says how much. This says to whom, and it is decided on its own. Under Who it may pay in Settings: - Any listed service. The agent may use services from the public catalogue, with the checks above. - Only services I approve. A service that is not on your list asks you first, whatever it costs, even one cent inside the budget. The question names the service and the address it is paid at. Yo","s":"A budget says how much. This says to whom, and it is decided on its own. Under Who it may pay in Settings: - Any listed service. The agent may use services from the public catalogue, with the checks above. - Only services I approve. A service that is not on your list asks you first, whatever it costs, even one cent inside the budget. The question names the service and the address it is paid at. Your yes approves the service for this and later requests. It does not approve a purchase: each one still has to fit your limits. The list is yours: approve or block a service by its address in Settings, and take it off again. A blocked service is never paid and never shown to the agent. In both settings, a service that starts asking to be paid at a different address than the one you approved, or the one it was paid at before, is asked about again. Looking through the services Under Look through services that can be paid , search the public catalogue by words (\"web search\", \"weather\"). Each result shows what the service says it does, the price of a call and how many others have paid it in the last 30 days, with Approve and Block . Looking pays nothing. The catalogue is a public list anyone can add to, so a description is the service's own claim. A team's rule In a team, the owner and admins can decide for everyone. Under Your team's rule : - Members decide. Each member's own settings apply, with the team's approved and blocked services added. - Only services approved here. The team's agents pay the services on the team's list and no others. A member is not asked about a service that is missing, cannot add one, and their own list does not override the team's. What the team blocked is never paid. How much stays each member's own: their limits, their wallet. The rule is kept with the team at Graspable and checked by the agent on each member's computer before anything is signed. It covers what Graspable's agent buys. It is not custody: a member's wallet is still theirs, and the rule does not stop what they do with it outside Graspable. Graspable checks this before anything is signed. The AI model cannot change the list, no judge answers for you, and when nobody can be asked, a service that is not approved is not paid. That holds with \"spend on its own\" switched on too. When nobody can be asked (the request runs without asking, or the agent is the Codex CLI, which cannot put a question to you), a purchase that needs your yes is refused.","u":"/docs/agent-spending#who-the-agent-may-pay"},{"k":"docs","t":"Agent spending","h":"Spending on its own (experimental, unsafe)","x":"Let the agent spend on its own, without asking removes the questions. The setting above decides instead: Finish fast pays even through doubt, Balanced pays when the judge agrees or when there is no judge, Save money pays only when the judge is sure. An AI model can be wrong, and text it reads on the web can try to talk it into buying. With this on, the limits are the only thing between the agent a","s":"Let the agent spend on its own, without asking removes the questions. The setting above decides instead: Finish fast pays even through doubt, Balanced pays when the judge agrees or when there is no judge, Save money pays only when the judge is sure. An AI model can be wrong, and text it reads on the web can try to talk it into buying. With this on, the limits are the only thing between the agent and the wallet. Assume that everything inside them can be spent, and keep in the wallet only what you accept losing. Three things hold even then: nothing above the limit for one purchase is bought, a purchase the judge traces to text the agent read is refused, and the agent buys only from services that at least three others have paid in the last 30 days.","u":"/docs/agent-spending#spending-on-its-own-experimental-unsafe"},{"k":"docs","t":"Agent spending","h":"The record","x":"Settings → Agent purchases is a page of its own for following the spending. It shows what was paid in the period you choose, per day, by service, by project, by agent and by who decided, and lists every purchase and every refusal; you can narrow it to a project, a service, or only what was paid. Open one to see: - the request it was for, and the agent's own reason; - what else it considered; - who","s":"Settings → Agent purchases is a page of its own for following the spending. It shows what was paid in the period you choose, per day, by service, by project, by agent and by who decided, and lists every purchase and every refusal; you can narrow it to a project, a service, or only what was paid. Open one to see: - the request it was for, and the agent's own reason; - what else it considered; - who decided (a limit, the judge, your setting, or you), and why; - the judge's answers with their scores, and which judge it was; - the budget at that moment, the service, and the payment itself. CSV and JSON save the record as a file. Each purchase also appears in the request's transcript as it happens. The record is a file on your computer ( ~/.graspable/payments ). It is not sent to Graspable, and neither is anything else about a purchase: not the amount, not the service, not that one happened.","u":"/docs/agent-spending#the-record"},{"k":"docs","t":"Agent spending","h":"What the judge is shown","x":"Your request, the agent's reason, what the service says of itself, and its price. Never the wallet, its balance, the receiver or a transaction. Under Who judges whether a purchase is worth it you choose: System One as you set it up, only your own System One providers, or nobody (you are then asked whenever a judge would be needed). With the first choice and no provider of your own, the judgement p","s":"Your request, the agent's reason, what the service says of itself, and its price. Never the wallet, its balance, the receiver or a transaction. Under Who judges whether a purchase is worth it you choose: System One as you set it up, only your own System One providers, or nobody (you are then asked whenever a judge would be needed). With the first choice and no provider of your own, the judgement passes through Graspable's System One like its other judgements. Choose the second or third to keep purchases entirely away from Graspable.","u":"/docs/agent-spending#what-the-judge-is-shown"},{"k":"docs","t":"Agent spending","h":"Limits of this first version","x":"- It pays in USDC on Base only, to services that speak the x402 protocol. - Prices must be fixed per call. Services with open-ended prices are not used. - It has been tried against real services up to the point of payment. Treat it as new.","u":"/docs/agent-spending#limits-of-this-first-version"},{"k":"docs","t":"Native XR","x":"Build native XR with Graspable's agent in Unity, Unreal or Swift and RealityKit. Choose a project type, prepare its tools and test on your device. Graspable is a desktop AI app builder. Its agent edits ordinary project files: React Three Fiber and WebXR for browser experiences, Unity or Unreal Engine for native projects, and Swift with RealityKit for its Apple Vision Pro template. You can open tho","s":"Build native XR with Graspable's agent in Unity, Unreal or Swift and RealityKit. Choose a project type, prepare its tools and test on your device. Graspable is a desktop AI app builder. Its agent edits ordinary project files: React Three Fiber and WebXR for browser experiences, Unity or Unreal Engine for native projects, and Swift with RealityKit for its Apple Vision Pro template. You can open those files in the corresponding editor and keep working by hand. Choose a native project when you need a device SDK, installed-app distribution or features your target browser cannot provide. WebXR projects open from a link in supported browsers; immersive modes and input depend on the device. A native template still needs its engine, build modules, device permissions and signing where required. Graspable uses those tools on your computer to build and inspect the result. The native templates (Unity, Unreal Engine, Apple Vision Pro) and the plugins that drive those engines come with the Pro plan and higher; the Expo panel app comes with Builder. They run on your computer only: a Cloud Agent cannot build them, because the engine is on your disk and signed in as you.","u":"/docs/native-xr"},{"k":"docs","t":"Native XR","h":"Can one project target every headset?","x":"Each template creates a project for its chosen stack. The Unity headset templates share a starting scaffold, but use different device packages and settings. The Apple Vision Pro template creates a separate SwiftUI and RealityKit project. Graspable does not automatically convert a WebXR or Unity project into that Swift project. For example, a native Quest app and a native Vision Pro app made from t","s":"Each template creates a project for its chosen stack. The Unity headset templates share a starting scaffold, but use different device packages and settings. The Apple Vision Pro template creates a separate SwiftUI and RealityKit project. Graspable does not automatically convert a WebXR or Unity project into that Swift project. For example, a native Quest app and a native Vision Pro app made from these templates are two projects. You may reuse compatible models, textures and parts of the design, but interaction code, build configuration and device tests belong to each target. For a shared web project, check whether every intended browser supports the immersive features you use.","u":"/docs/native-xr#can-one-project-target-every-headset"},{"k":"docs","t":"Native XR","h":"Should I use Graspable or Unity?","x":"You can use both: choose a Unity template in Graspable to have its agent work on a Unity project. Unity remains the engine that compiles and runs it. Its installation, licensing, target modules and device setup still apply. Use the Unity Editor for manual scene editing and its own development tools; use Graspable for agent edits and the checks described below. If you want a browser experience inst","s":"You can use both: choose a Unity template in Graspable to have its agent work on a Unity project. Unity remains the engine that compiles and runs it. Its installation, licensing, target modules and device setup still apply. Use the Unity Editor for manual scene editing and its own development tools; use Graspable for agent edits and the checks described below. If you want a browser experience instead, start with React XR. That creates a web project, with a different rendering and interaction stack. See the Unity comparison for the workflow choices, and AR films and tabletop games for phone-browser tracking limits.","u":"/docs/native-xr#should-i-use-graspable-or-unity"},{"k":"docs","t":"Native XR","h":"What you need","x":"For the Unity templates - Unity 6 installed with Unity Hub, with the Android module (Unity Hub → the Editor's settings → Add modules → Android Build Support, with SDK, NDK and OpenJDK). - The Unity CLI , Unity's own command-line tool. It is what Graspable uses to drive the Editor without a window. - A Unity account . Unity Personal is free below Unity's revenue threshold; the Editor checks its lic","s":"For the Unity templates - Unity 6 installed with Unity Hub, with the Android module (Unity Hub → the Editor's settings → Add modules → Android Build Support, with SDK, NDK and OpenJDK). - The Unity CLI , Unity's own command-line tool. It is what Graspable uses to drive the Editor without a window. - A Unity account . Unity Personal is free below Unity's revenue threshold; the Editor checks its licence through Unity Hub. For the Apple Vision Pro template - A Mac with Xcode and its visionOS platform (Xcode → Settings → Components). The visionOS Simulator comes with it. For the Unreal Engine template - Unreal Engine 5 , installed with the Epic Games Launcher. On a Mac also Xcode with its Metal Toolchain component. With 5.8 or later the agent also writes Blueprints (experimental). - For C++ in an Unreal project, the compiler your engine asks for: Xcode on a Mac, Visual Studio 2022 with the workload “Game development with C++” on Windows; on Linux the engine brings its own. Settings → Native engines shows what your engine needs and whether it is there, and has a button that opens the App Store's Xcode page on a Mac. On Windows the button starts Visual Studio's installer through Windows' own package manager; that path is new and has been tried less than the Mac one, so if it does not finish, install Visual Studio 2022 with that workload yourself and Graspable finds it. Settings → Native engines shows what Graspable found: the Unity Editors and their modules, the Unity CLI, Xcode and its visionOS simulators, and what is missing. Graspable installs none of it and never signs in to Unity or Apple for you; it uses what is on your computer.","u":"/docs/native-xr#what-you-need"},{"k":"docs","t":"Native XR","h":"The Unity templates","x":"Three headset templates share a Unity scaffold and differ in their target packages and settings: Unity for Quest (Meta Quest 3, Quest 3S, Meta VR Glasses, with Unity's Meta OpenXR package), Unity for Android XR (Android XR headsets and glasses, with Unity's Android XR package) and Unity for XREAL (XREAL glasses, through Android XR). Each creates its own project with Unity 6, OpenXR, AR Foundation ","s":"Three headset templates share a Unity scaffold and differ in their target packages and settings: Unity for Quest (Meta Quest 3, Quest 3S, Meta VR Glasses, with Unity's Meta OpenXR package), Unity for Android XR (Android XR headsets and glasses, with Unity's Android XR package) and Unity for XREAL (XREAL glasses, through Android XR). Each creates its own project with Unity 6, OpenXR, AR Foundation for the real room, the XR Interaction Toolkit 3 with XR Hands, the Universal Render Pipeline, and Unity's Pipeline package, which lets Graspable talk to the Editor. The start scene is mixed reality. The real room shows through, a small table holds a ball you can grab and a button you can press, and you place a marker on any real surface by pointing and pinching. Hands, gaze and controllers all work; Meta VR Glasses have no controllers, so the scene is designed for hands first. The scene's rig is Unity's own for hands and controllers together, and it switches by itself: bare hands are shown, poke with a fingertip and pinch near or far; controllers point, grab, select, move, turn and teleport; gaze works where the headset tracks eyes. Everything the agent adds works with all of them. The kit under Assets/Graspable is what the agent builds with: - Materials that look like what they are. Photographed concrete, plaster, oak, walnut, steel plate, painted metal and rubber, with their roughness and relief, and plain metals, paints and plastics. Boxes keep the texture at its real size, so a long board does not stretch its grain. - Models. Shelves, a tool chest, a cardboard box and a drill are included, and any glTF model you add is imported with its materials. - Behaviour. Labels sized in metres so they stay readable, scene buttons, grabbable things, placing on real surfaces with anchors, a hint panel that follows you, mechanisms that a hand moves (levers, doors, handwheels on hinges; drawers and faders on rails), and panels of controls in the scene (buttons, sliders, switches). The toolkit's own pieces are used as they are: sockets that hold a tool, two-handed scaling, teleport areas and anchors, climbing. The textures and models are from Poly Haven and are in the public domain (CC0). A room in the Editor On a headset the app is seen in your real room. In the Editor there is no room, so the templates switch on AR Foundation's XR Simulation with a workshop of Graspable's own: a concrete floor, plaster walls, an oak workbench, shelves. In play mode the camera shows that workshop where the headset would show your room, and its floor, walls and bench are found as surfaces, so placing things on surfaces, anchors and raycasts can be tried without a headset. The workshop is not part of your app and is never in a build. The agent works in the Editor The agent does not write scene files by hand. It reads the scene as the Editor sees it and changes it there: it creates and arranges objects, adds components and sets their properties, makes prefabs, sets materials, and can run a few lines of C in the Editor for what the commands cannot do. These are Unity's own commands, answered by the windowless Editor that also takes the pictures. Every change is saved to the project at once and asks for your approval the same way a file edit does.","u":"/docs/native-xr#the-unity-templates"},{"k":"docs","t":"Native XR","h":"Phone AR and tabletop films","x":"The Unity AR tabletop template starts with Signal Garden, an editable rescue film and relay puzzle. It uses AR Foundation for native iPhone and Android surface placement, and a separate OpenXR configuration for Quest passthrough. Choose Signal Garden → Target → Phone AR or Quest in Unity before building. Graspable's Build Android app action exports the selected configuration without automatically ","s":"The Unity AR tabletop template starts with Signal Garden, an editable rescue film and relay puzzle. It uses AR Foundation for native iPhone and Android surface placement, and a separate OpenXR configuration for Quest passthrough. Choose Signal Garden → Target → Phone AR or Quest in Unity before building. Graspable's Build Android app action exports the selected configuration without automatically installing it on a connected device. iOS needs Unity's iOS build module and Xcode signing. See AR films and tabletop games for the browser version and tracking requirements.","u":"/docs/native-xr#phone-ar-and-tabletop-films"},{"k":"docs","t":"Native XR","h":"The Apple Vision Pro template","x":"A native visionOS app in Swift: SwiftUI for a window with controls, RealityKit for an immersive space around you. The start app is a window with a finish picker (oak, painted metal, steel plate) and a button that opens the space, where a block of that material spins and can be looked at, pinched and dragged. The template has the same photographed surfaces as the Unity kit, as physically based mate","s":"A native visionOS app in Swift: SwiftUI for a window with controls, RealityKit for an immersive space around you. The start app is a window with a finish picker (oak, painted metal, steel plate) and a button that opens the space, where a block of that material spins and can be looked at, pinched and dragged. The template has the same photographed surfaces as the Unity kit, as physically based materials that the room you are in lights. The Xcode project is set up so that a new Swift file is part of the build the moment it exists; nobody edits the project file. The simulator shows the window in a virtual room. Real surfaces and hand tracking work on the device only, and the template says so to the agent, so what it builds stays usable in the simulator.","u":"/docs/native-xr#the-apple-vision-pro-template"},{"k":"docs","t":"Native XR","h":"The Unreal Engine template","x":"An Unreal Engine 5 project for OpenXR headsets, with one difference from an ordinary one: its level is a Python script. Unreal keeps levels and materials in binary files that nobody can read, review or have an agent edit as text. In this template Scripts/build.py describes the scene with a small kit (photographed materials, boxes and cylinders placed in metres, labels, light), and your own Unreal ","s":"An Unreal Engine 5 project for OpenXR headsets, with one difference from an ordinary one: its level is a Python script. Unreal keeps levels and materials in binary files that nobody can read, review or have an agent edit as text. In this template Scripts/build.py describes the scene with a small kit (photographed materials, boxes and cylinders placed in metres, labels, light), and your own Unreal Editor builds the level from it without a window. The agent edits the script; you can read every change. What it can and cannot do today: - A scene that looks like visualisation work : scanned furniture, props and plants at their real size, photographed surfaces that keep their scale on any shape, glass, daylight in real units with a sky seen through windows, spot and area lights, and a camera-style exposure. The models, surfaces and skies come from Poly Haven (free for any use) and are fetched by name the first time a scene uses them, then kept in the project. - Unreal's high renderer on the desktop : the picture, and PC VR, use deferred shading with light that bounces (Lumen) and its reflections. A standalone headset always uses Unreal's mobile renderer, without bounce; the agent keeps a little fill light so the scene reads there too. - A budget with every build : triangles, meshes, materials and lights, against what a standalone headset draws comfortably. The agent reads it and tells you. - Conventions a studio would recognise : Blueprints, materials and folders are named and placed the way most Unreal teams do it, so the project is readable when you open it in the Unreal Editor. The skills that carry these rules come with the template; see Skills. - The scene : materials, lights, rooms, layout and text are built from the script, and the agent sees a picture of the game from where your eyes would start. - The person : Unreal's own VR rig is the player, with a tracked head, two controllers or bare hands, grabbing with the grip or a pinch, teleporting and turning. Graspable copies the rig and the engine's headset settings from your Unreal Engine installation into the project on the first build. - Things to do : the script makes shapes that can be picked up and thrown, and parts held by real joints that you grab and move: doors, lids, levers and handwheels on hinges, drawers and faders on rails. It marks the floor you may teleport on. - Behaviour, in Blueprints : logic that reacts, such as a lever that switches a lamp or a button that starts something, lives in Blueprints. With Unreal Engine 5.8 or later the agent writes them itself; see The Unreal Editor, live. With an older engine a script cannot write that logic, and the agent says which reaction is missing instead of pretending. - Systems, in C++ : when a project needs a system of its own, such as a component many actors share or work that must be fast, the agent writes it as a C++ module in Source/ , following the template's skill for it. The check compiles it with the engine's own build tool before it builds the level, and a compiler error comes back with its file and line. A project without C++ stays as it is and needs no compiler. - Your own models : a model file you added to the project (glb, glTF, FBX, OBJ) is placed in the level by its path, like a scanned one by its name. - Build for headset packages the Android app with Unreal's own build tool and installs it over USB. It needs Unreal's Android set-up (the SDK, NDK and JDK versions of your engine, installed with the engine's SetupAndroid script); without it the button says so in a few seconds. You need Unreal Engine 5 on your computer. On a Mac, Unreal also needs Xcode's Metal Toolchain to draw (Xcode → Settings → Components). The first picture on a computer takes many minutes while Unreal prepares its shaders; a game window opens for a moment each time. Close the project in the Unreal Editor before a picture is taken: while an Editor has the project open, the game shows grey checkers instead of materials. The Unreal Editor, live This is experimental: it needs Unreal Engine 5.8 and has been tried on a Mac only. Unreal Engine 5.8 and later carry Epic's own server for AI agents inside the Editor. Graspable starts your Editor without a window, with that server on a port of this computer only, and gives the agent three tools: Tool What it does Asks first --- --- --- unreal look Lists the Editor's toolsets and their tools, and reads: a Blueprint's graph as text, its variables, the assets in a folder No unreal edit Makes and changes Blueprints, materials, UI widgets and data assets. A Blueprint's graph is written as a short text script and compiled As a file edit unreal run Runs a small script that calls several of those tools in one go As a command The level stays the Python script. A Blueprint holds the behaviour, is kept under Content/Blueprints , and the script places it in the level. The first call starts the Editor, which takes about half a minute once the engine's caches exist; it closes by itself after ten idle minutes. Nothing is switched on in your project file: the server is enabled for that start only. UE-MCP, an optional plugin Experimental, like the tools above. UE-MCP is an open-source set of Editor tools made by its own community: Blueprints, materials, levels, animation, audio, effects, UI and more. Switch it on under Settings → Plugins and connectors . UE-MCP reaches the Editor through a bridge that must be compiled for your engine. Graspable does that once for each Unreal Engine version, in the background, on the first request on an Unreal project: it takes several minutes, the request carries on without the tools and says so, and a later request has them. After that the compiled bridge, about 20 MB, is copied into each Unreal project the agent works on. - It needs Unreal Engine 5.6 or later (5.4 on Windows) and a C++ compiler: Xcode on a Mac, Visual Studio with the game development workload on Windows. - With Unreal Engine 5.8 the tools work in the Editor Graspable runs without a window. With an older engine they work while the project is open in the Unreal Editor. - The bridge lives in the project's Plugins folder. It is left out of the project's history and of what you publish. - Graspable uses a fixed version of UE-MCP, which changes only with a Graspable update, and leaves out its tool for filing reports with its authors. - The agent asks before it uses one of its tools, as with every plugin. See Plugins and connectors.","u":"/docs/native-xr#the-unreal-engine-template"},{"k":"docs","t":"Native XR","h":"The Panel app template","x":"A flat app that opens as a floating window in a headset's home space. Meta Quest, Meta VR Glasses and Android XR run Android apps this way, and a panel is often the right shape for a tool. It is an Expo project (React Native): it runs as a web page in the preview, a Cloud Agent can work on it, and Expo's own tools build it as an Android app for the headsets.","u":"/docs/native-xr#the-panel-app-template"},{"k":"docs","t":"Native XR","h":"How the agent works in a native project","x":"The agent edits files: C under Assets/Graspable/Scripts and scenes under Assets/Scenes in a Unity project; Swift under App/ in a Vision Pro project. It never writes Unity's .meta files or the Xcode project file. When the agent wants to see its work, it calls verify app , as it does for a web project. For a native project that means: 1. Build. Unity: the Editor opens the project in batch mode and c","s":"The agent edits files: C under Assets/Graspable/Scripts and scenes under Assets/Scenes in a Unity project; Swift under App/ in a Vision Pro project. It never writes Unity's .meta files or the Xcode project file. When the agent wants to see its work, it calls verify app , as it does for a web project. For a native project that means: 1. Build. Unity: the Editor opens the project in batch mode and compiles every script; errors come back as File.cs(line): CS1234: message . Vision Pro: Xcode builds the app for the simulator; errors come back as App/File.swift:line:col: message . The agent fixes them first. 2. Run. Unity: an Editor stays open for the project without a window, enters play mode in the simulated workshop, lets the app run for a moment (the agent can ask for a later moment) and renders the main camera to a picture, with the console. Vision Pro: the app is installed and launched in the visionOS Simulator with its immersive space opened, and after a moment the simulator's screen is saved. 3. Report. The agent gets the picture and the errors. A run ends only when the project builds and runs without errors; failures go back to the agent for repair, as always.","u":"/docs/native-xr#how-the-agent-works-in-a-native-project"},{"k":"docs","t":"Native XR","h":"The preview","x":"A native project has no dev server. The preview pane shows the picture the engine took; Capture a frame (Unity) or Run in the simulator (Vision Pro) takes a new one. Open in Unity and Open in Xcode open the project in your own editor window. One Unity Editor runs per project: opening the window takes over from the windowless one, which is also closed after ten minutes without use and when Graspabl","s":"A native project has no dev server. The preview pane shows the picture the engine took; Capture a frame (Unity) or Run in the simulator (Vision Pro) takes a new one. Open in Unity and Open in Xcode open the project in your own editor window. One Unity Editor runs per project: opening the window takes over from the windowless one, which is also closed after ten minutes without use and when Graspable quits. It holds a Unity licence seat while it runs, like any open Editor. The previous picture stays visible while you capture another frame, including if the new capture fails. Replacing the picture or closing the preview releases the old image from memory.","u":"/docs/native-xr#the-preview"},{"k":"docs","t":"Native XR","h":"Getting the app onto a headset","x":"Unity. Build for headset in the preview pane builds the Android app and, when a headset is plugged in over USB with developer mode on, installs it there; the app then appears under Unknown Sources in the headset's library. Without a headset the app lands in the project's Builds/ folder, for Meta Quest Developer Hub or adb install . The first build takes many minutes; later ones are shorter. Buildi","s":"Unity. Build for headset in the preview pane builds the Android app and, when a headset is plugged in over USB with developer mode on, installs it there; the app then appears under Unknown Sources in the headset's library. Without a headset the app lands in the project's Builds/ folder, for Meta Quest Developer Hub or adb install . The first build takes many minutes; later ones are shorter. Building needs the Android module in Unity; installing needs the Android platform tools ( adb ) on your computer. Apple Vision Pro. Open the project in Xcode, choose your developer account under Signing, and run it on your Vision Pro. Panel app. npm run android in the project builds and installs the Android app with Expo.","u":"/docs/native-xr#getting-the-app-onto-a-headset"},{"k":"docs","t":"Native XR","h":"Publishing and examples","x":"Publish works for native projects too: the address shows the picture from the project's check, others can continue from its source, and a Unity or Unreal Engine project can offer its headset app as a download and link to its pages in the stores. See Publishing. Made in Graspable has an example for each native template, each built from one request that you can read and continue from.","u":"/docs/native-xr#publishing-and-examples"},{"k":"docs","t":"Native XR","h":"What is not there yet","x":"- Unreal Engine on a headset: the template builds and pictures the level on your computer; its rig, grabbing and joints are Unreal's own and are set up by the script, but have not been tried on a device. Passthrough in Unreal needs the headset maker's plugin, which the template does not include. - Meta's own simulator and Meta XR Operator: Graspable's play-mode check uses AR Foundation's XR Simula","s":"- Unreal Engine on a headset: the template builds and pictures the level on your computer; its rig, grabbing and joints are Unreal's own and are set up by the script, but have not been tried on a device. Passthrough in Unreal needs the headset maker's plugin, which the template does not include. - Meta's own simulator and Meta XR Operator: Graspable's play-mode check uses AR Foundation's XR Simulation. You can use Meta's tools in the Unity window as you would without Graspable. - Emulators for Android XR and XREAL: the check runs play mode in the Editor's simulated room; the headset is where you try the real thing. - Hands in the picture: the simulated room has surfaces and a camera, but nobody pinching. What needs a pinch is tried on the headset, or shown by a demo mode you ask the agent for. - Recorded headset sessions as tests (fixtures) exist for web projects only.","u":"/docs/native-xr#what-is-not-there-yet"},{"k":"docs","t":"Native XR","h":"Where things are","x":"Web project Unity project Vision Pro project --- --- --- --- Code App.jsx , components/ Assets/Graspable/Scripts/ .cs , Assets/Scenes/ .unity App/ .swift , App/Views/ Packages package.json Packages/manifest.json none (Apple's frameworks) First install npm install the Editor's first open (resolves packages, compiles) the first Xcode build Check build, load in a headless browser, replay fixtures com","s":"Web project Unity project Vision Pro project --- --- --- --- Code App.jsx , components/ Assets/Graspable/Scripts/ .cs , Assets/Scenes/ .unity App/ .swift , App/Views/ Packages package.json Packages/manifest.json none (Apple's frameworks) First install npm install the Editor's first open (resolves packages, compiles) the first Xcode build Check build, load in a headless browser, replay fixtures compile, a frame of play mode build, run in the simulator Preview the dev server in the app, with an emulated headset a rendered frame; or the Unity window the simulator's screen; or Xcode Hidden from the file list node modules , dist Library , Temp , Logs , Builds , .meta files the .xcodeproj , derived data","u":"/docs/native-xr#where-things-are"},{"k":"docs","t":"Meta glasses","x":"Build for Ray-Ban Meta, Oakley Meta and Meta Ray-Ban Display glasses: a phone app on Meta's Device Access Toolkit, or a web app for the Display glasses. Choose the glasses and the interaction before choosing a template. For a camera-based app on Ray-Ban Meta or Oakley Meta glasses, start with a phone companion built on Meta's Wearables Device Access Toolkit. For a small visual app on Meta Ray-Ban ","s":"Build for Ray-Ban Meta, Oakley Meta and Meta Ray-Ban Display glasses: a phone app on Meta's Device Access Toolkit, or a web app for the Display glasses. Choose the glasses and the interaction before choosing a template. For a camera-based app on Ray-Ban Meta or Oakley Meta glasses, start with a phone companion built on Meta's Wearables Device Access Toolkit. For a small visual app on Meta Ray-Ban Display, Graspable also offers a standalone web template. Meta documents both development paths.","u":"/docs/meta-glasses"},{"k":"docs","t":"Meta glasses","h":"Choose your starting point","x":"What you want to build Start here Check before you commit --- --- --- A hands-free photo journal with spoken feedback Glasses field notes Some inputs and speech features use experimental toolkit capabilities. The app must handle those being unavailable. A phone companion that uses the glasses' camera or sends a display screen Meta glasses app Your target glasses must support each capability. Nativ","s":"What you want to build Start here Check before you commit --- --- --- A hands-free photo journal with spoken feedback Glasses field notes Some inputs and speech features use experimental toolkit capabilities. The app must handle those being unavailable. A phone companion that uses the glasses' camera or sends a display screen Meta glasses app Your target glasses must support each capability. Native phone builds need the platform's development tools. A timer, checklist or short instructions on Meta Ray-Ban Display Display glasses web app The web path has different hardware access from the phone toolkit. Publishing the URL does not grant access to every glasses feature. A room-scale scene with 3D objects, gaze and hand interactions The WebXR guide for Meta VR Glasses Meta VR Glasses are a separate development target from Meta's everyday AI glasses. For Android XR AI glasses, Google documents Jetpack Compose Glimmer. Its guidance covers both display and audio-only experiences. The Meta templates on this page do not establish compatibility with Android XR glasses; choose that platform's tools and validate its device support separately.","u":"/docs/meta-glasses#choose-your-starting-point"},{"k":"docs","t":"Meta glasses","h":"Compare the Meta development paths","x":"Meta glasses app Display glasses web app --- --- --- Glasses Ray-Ban Meta Gen 1 and Gen 2, Ray-Ban Meta Optics, Oakley Meta HSTN and Vanguard, Meta Ray-Ban Display Meta Ray-Ban Display only What it is An Expo app on your phone that talks to the glasses through the Meta AI app A web page the glasses show, 600 by 600 pixels Can use The camera stream and photos, and the Display screen; inputs, motion","s":"Meta glasses app Display glasses web app --- --- --- Glasses Ray-Ban Meta Gen 1 and Gen 2, Ray-Ban Meta Optics, Oakley Meta HSTN and Vanguard, Meta Ray-Ban Display Meta Ray-Ban Display only What it is An Expo app on your phone that talks to the glasses through the Meta AI app A web page the glasses show, 600 by 600 pixels Can use The camera stream and photos, and the Display screen; inputs, motion, speech and voice launch have experimental restrictions described below Motion and orientation, location, text by voice or handwriting, speech out, storage Cannot use The camera and the microphone Runs On a phone, built with Xcode or Android Studio From any https address; publishing from Graspable gives it one Both come with the Builder plan and higher. The native route builds and installs a companion app on your phone. The web route publishes an HTTPS address that you add to the Display glasses. Check Meta's current release requirements before promising a feature to customers: a working developer-mode demo does not establish public distribution eligibility.","u":"/docs/meta-glasses#compare-the-meta-development-paths"},{"k":"docs","t":"Meta glasses","h":"Before you start","x":"- Install the Meta AI app on your phone and connect the glasses in it. - Turn on Developer Mode in the Meta AI app: Settings, App Info, tap the version number five times, then confirm. With it on, an unpublished app may register with your glasses. - For the phone app: an iPhone on iOS 17.2 or later and a Mac with Xcode 26.4 or later, following Meta's current setup requirements. The Android templat","s":"- Install the Meta AI app on your phone and connect the glasses in it. - Turn on Developer Mode in the Meta AI app: Settings, App Info, tap the version number five times, then confirm. With it on, an unpublished app may register with your glasses. - For the phone app: an iPhone on iOS 17.2 or later and a Mac with Xcode 26.4 or later, following Meta's current setup requirements. The Android template needs Android 12 or later with Android Studio; its Expo wrapper has a higher minimum than Meta's SDK alone. - Match the Meta AI app and glasses software to the toolkit version in your project. Meta's compatibility table lists Meta AI 290 for toolkit 1.0.0 and 292 for 1.0.1, with glasses software 128 for both. An iOS dependency update can change the toolkit patch version. Check the table again after updating dependencies. Matching versions does not grant every capability. Meta can also enable features per account, so a device may decline an input, speech or display request even when its software is current. Handle that outcome in the app and test on the glasses you intend to support.","u":"/docs/meta-glasses#before-you-start"},{"k":"docs","t":"Meta glasses","h":"The phone app","x":"Create a project from Meta glasses app or Glasses field notes and describe what you want. The preview runs the app against the glasses emulator, described below, so you can build and try everything before it goes near a phone. To run it on your iPhone with your glasses, press Run on my phone above the preview. Run on my phone Plug the iPhone into the Mac with a cable the first time, unlock it and ","s":"Create a project from Meta glasses app or Glasses field notes and describe what you want. The preview runs the app against the glasses emulator, described below, so you can build and try everything before it goes near a phone. To run it on your iPhone with your glasses, press Run on my phone above the preview. Run on my phone Plug the iPhone into the Mac with a cable the first time, unlock it and tap Trust. Graspable builds the app with your Xcode, installs it and opens it. The first build takes several minutes. After that the app on the phone loads its code from your computer, so each change the agent makes shows on the phone in a few seconds, and pressing the button again takes about ten seconds. A new build is made only when the app's settings or its native packages change. What it needs, and the bar tells you which is missing: - For an iPhone: a Mac with Xcode from the App Store, opened once, and CocoaPods. For an Android phone: Android Studio, opened once, on a Mac, Windows or Linux computer. - Your Apple ID added in Xcode under Settings, Accounts. Meta's toolkit needs Wi-Fi access that Apple lets only a paid Apple Developer membership sign, so a free account cannot build these apps. - Developer Mode on the iPhone: Settings, Privacy & Security, Developer Mode. - The phone awake and on the same Wi-Fi as the computer. The first time, the phone asks whether the app may find devices on your network. Allow it. While it is on, the project's code is offered to devices on your network, as with any Expo development server. Stop ends that. The first time you run a project, Graspable gives it an app identifier of its own in app.json , because Apple signs each identifier for one developer. The same build works without glasses: the Mock screen, and \"Try with pretend ones\" in Field Notes, pair Meta's pretend glasses inside the app on the phone. An Android phone works the same way with Android Studio installed and opened once (it brings the Android SDK and Java) and USB debugging allowed on the phone. The connection then runs over the cable, so the phone need not be on the Wi-Fi. On the phone, the Glasses screen registers the app with Meta AI (the Meta AI app opens and asks you), asks for the camera and microphone, lists your glasses with their battery, wear and thermal state, and opens a session. The other screens use it: Camera streams the glasses' camera into the phone and takes photos; Display sends a screen to Meta Ray-Ban Display glasses and gets the taps back; Senses shows inputs, motion and speech; Extras lets the glasses open the app by voice and take a full-quality photo; Mock pairs pretend glasses of every model so you can try all of it without hardware. What the toolkit calls experimental (inputs, motion, speech, voice launch, standalone photos, audio in the stream) runs with Developer Mode on, and an app that uses it cannot be published to other people until Meta opens it. The camera stream, photos from it and the display can be published through Meta's own review and release channels, from the Wearables Developer Center. Field Notes, a showcase Glasses field notes is the same toolkit as one finished app for glasses without a display. Press Look and the phone shows what the glasses see. The shutter, or a tap on the temple, saves a moment. Say a note listens through the glasses and writes the first sentence under the photo. The phone answers \"saved\" and \"noted\" through the glasses' speakers, which are your phone's audio output while you wear them. Temple taps, the horizon line and spoken notes use the experimental parts of the toolkit; where the glasses refuse one, the app carries on without it. In the preview, the emulator stands in for the glasses.","u":"/docs/meta-glasses#the-phone-app"},{"k":"docs","t":"Meta glasses","h":"The glasses emulator","x":"The preview of a phone app shows it in a phone-sized frame with a panel beside it: pretend glasses and a pretend Meta AI app. Nothing needs installing, and it works on any computer. - The Meta AI app. When the app registers or asks for the camera or the microphone, a sheet slides over the phone as the Meta AI app would, and you answer it. A switch turns Developer Mode off, and registering then fai","s":"The preview of a phone app shows it in a phone-sized frame with a panel beside it: pretend glasses and a pretend Meta AI app. Nothing needs installing, and it works on any computer. - The Meta AI app. When the app registers or asks for the camera or the microphone, a sheet slides over the phone as the Meta AI app would, and you answer it. A switch turns Developer Mode off, and registering then fails the way it does on a real phone. - The glasses. A pair of Ray-Ban Meta to start with; add any model. Switch a pair off, take it out of range, take it off, fold it, drain its battery, overheat it, or make it need an update, and watch what your app does. - What the glasses see. A test card, a camera of this computer (a phone used as a webcam works too), the screen, or a picture or video file. A slider tilts the wearer's head. - On the glasses. Tap the temple, swipe, press the capture or action button. - Voice. Type a sentence and the glasses \"hear\" it, or speak if the computer has a microphone. The panel shows what your app said back into the wearer's ear. - What the glasses hear. For an app that asks for sound with the camera stream: silence, a steady tone, a microphone of this computer, or a sound file. Your app gets it in the form the glasses send it, and a meter shows the level. - The display. With a Meta Ray-Ban Display pair added, the panel draws the screen your app sends, and its buttons can be pressed. - Break something. Make registering, a permission, the stream or a photo fail on purpose. The agent uses the same emulator to check its work. After a change it can pair the glasses, answer the Meta AI app, start the camera, tap the temple, say a sentence, fold the glasses or make a step fail, and read what your app showed and said. Ask it to \"try what happens when the glasses overheat\" and it does that. The first time a project uses a camera or a microphone of your computer, your system asks whether Graspable may use it. The emulator follows Meta's toolkit closely, including the pause before glasses become ready, but it is not the glasses. Timing, picture quality, the microphones and what Meta's servers allow can differ. Try on a real pair before you rely on something.","u":"/docs/meta-glasses#the-glasses-emulator"},{"k":"docs","t":"Meta glasses","h":"The web app for Display glasses","x":"Create a project from Display glasses web app . The preview shows the app in a frame of the glasses' size: the arrow keys move, Enter selects, Escape goes back, which is what the Neural Band and the touchpad send. Publish the app as you publish any Graspable app. Its page then shows an install code beside the frame; scan it with the glasses or the Meta AI app, and the app appears at the end of the","s":"Create a project from Display glasses web app . The preview shows the app in a frame of the glasses' size: the arrow keys move, Enter selects, Escape goes back, which is what the Neural Band and the touchpad send. Publish the app as you publish any Graspable app. Its page then shows an install code beside the frame; scan it with the glasses or the Meta AI app, and the app appears at the end of the app grid on the glasses. Without the code: Meta AI app, Devices, the Display glasses, App connections, Web apps, Add a web app, then the address. Publishing is the hosting. A published app lives at its own https address on graspable.cloud, which is what the glasses need; there is nothing else to set up. Publish again and the glasses load the new version the next time the app opens. Once loaded, the app also opens without a connection. Voice. Each screen offers its actions to Meta AI as voice tools, so the wearer can say \"Hey Meta, set the timer to three minutes\" with the app open. Meta calls this WebMCP; it is a developer preview on the glasses and off by default, so every action also has a button. On a computer, the page lists the tools of the screen that is showing beside the frame, each with a Try button that calls the tool the way Meta AI would. Ask the agent for a new tool in words: \"let me ask for the next step by voice\". The display is additive, so black is transparent and bright colours show. Keep text large, screens short and one thing per screen. The agent knows these rules. The Meta Ray-Ban Display Simulator, a Chrome extension from Meta, shows a page the way the glasses do.","u":"/docs/meta-glasses#the-web-app-for-display-glasses"},{"k":"docs","t":"Meta glasses","h":"What Graspable does not do","x":"- It does not publish to Meta for you. Meta's review, release channels and testers are in the Wearables Developer Center, with your own Meta developer account. - It does not hide the capture LED, record people without their knowledge, or send what the camera sees anywhere you did not ask for, and the agent is told not to build such things. - A published Display web app is an ordinary web page at y","s":"- It does not publish to Meta for you. Meta's review, release channels and testers are in the Wearables Developer Center, with your own Meta developer account. - It does not hide the capture LED, record people without their knowledge, or send what the camera sees anywhere you did not ask for, and the agent is told not to build such things. - A published Display web app is an ordinary web page at your app's address; it holds no key and knows nothing of your Meta account.","u":"/docs/meta-glasses#what-graspable-does-not-do"},{"k":"docs","t":"Publish and share","x":"Put your app on the web at an address of its own, keep some apps private, and let other people continue from what you made. Every project can become an app on the web. Graspable builds it, hosts it and gives it an address of its own. People open it in a browser or in a headset; nothing needs to be installed.","u":"/docs/publishing"},{"k":"docs","t":"Publish and share","h":"Publish an app","x":"Open the project and choose Publish in the header. Give the app a title and, if you like, a line about it, then choose Publish . Graspable builds the project, uploads it and shows the address. Every app has a name , and the name is its address: an app named quiet-otter-482 is at https://quiet-otter-482.graspable.cloud . A new app gets a random name that is easy to say. To choose your own, open Set","s":"Open the project and choose Publish in the header. Give the app a title and, if you like, a line about it, then choose Publish . Graspable builds the project, uploads it and shows the address. Every app has a name , and the name is its address: an app named quiet-otter-482 is at https://quiet-otter-482.graspable.cloud . A new app gets a random name that is easy to say. To choose your own, open Settings → Published apps and choose Rename : 3 to 40 lowercase letters, digits and dashes, and not a name another app already has. Renaming changes the address; the old one stops working. Publishing again from the same project updates the app at the same address. From a terminal, grasp publish <project does the same, and --name asks for a name (see CLI). If a file upload loses its connection or the hosting service has a temporary error, Graspable retries that file up to twice. If publishing still fails, your previous published version stays available. Permission and size errors stop the upload immediately. Each app runs at its own address, apart from every other app and from graspable.dev, so it can use the headset but cannot reach your account or anyone else's app. Your published apps are listed under Settings → Published apps , each as a card with its picture, its views and likes and its address. Under Manage on a card you can rename the app, make it private or public, allow or stop people continuing from it, or take it down.","u":"/docs/publishing#publish-an-app"},{"k":"docs","t":"Publish and share","h":"Your handle","x":"A handle is optional. Choose one under Settings → Graspable account : 3 to 30 characters, lowercase letters, digits, and - . With a handle, your public apps are listed on your page, graspable.dev/@you , and each app's page says who made it.","u":"/docs/publishing#your-handle"},{"k":"docs","t":"Publish and share","h":"Community: views, likes, follows and collections","x":"A public app can be found by other people. Community in the app (the button beside Settings on the Projects page) lists public apps: the most viewed, the most liked, the newest, and those of the people you follow. On a card you can: - Open the app, or its page on graspable.dev. - Like it. One like per account; click again to take it back. - Follow the person who made it. What they publish next is ","s":"A public app can be found by other people. Community in the app (the button beside Settings on the Projects page) lists public apps: the most viewed, the most liked, the newest, and those of the people you follow. On a card you can: - Open the app, or its page on graspable.dev. - Like it. One like per account; click again to take it back. - Follow the person who made it. What they publish next is under \"People you follow\". - Save it to a collection. A collection is a named list of apps with a page of its own ( graspable.dev/c/… ) that you can share; it also appears on your handle's page. - Continue from it , when its maker allows that. Liking, following and saving need your Graspable account. Looking needs nothing. Being found: category and tags Under Settings → Published apps , a public app has a line \"In the gallery as\": choose one category (Games, Education, Training, Product, Architecture, Art, Tools, Social, Science or Other) and up to five tags, separated by commas. Tags are lower case, two to twenty-four characters, letters, digits and dashes. The gallery on graspable.dev/made lists every public app. People search it by name, tag or maker, filter it by category or by clicking a tag, and order it by trending, most viewed, most liked or newest. Trending is this week's views and likes, a like counting as five views. The home page shows the four apps most opened and liked this week, and the rest of the week's trending ones in a row beside them. Your app's page Every public app has a page on graspable.dev, at graspable.dev/a/<its id . The address never changes, whatever you rename the app to. The page shows the app's pictures and films, what it is about, who made it, where it runs, its views and likes, and buttons to open it, like it, save it to a collection, share it and, when you allow it, continue from it. You fill the page in on the website: sign in, open Account → Published apps , and choose Edit its page on an app. - Pictures: up to 8, PNG, JPEG or WebP, up to 5 MB each. A 16:9 picture fits best. Move them into the order you want; the first is shown large. - Films: up to 2, by the address of their page at YouTube or Vimeo. A film is loaded from that service only when a visitor starts it. - The picture in lists: choose Use in lists on one of your pictures, and the gallery and shared links show that one. Without a choice, it is the picture taken when you published. - About this app: up to 4,000 characters. Leave an empty line between paragraphs. - Where people send feedback: an https address, such as a form, a forum or your Discord. Without one, the page points people to your profile and its links. The title, the one line under it, the category and the tags are in the app's settings, under Published apps . With no pictures of your own, the page shows the picture taken when you published. Sharing a public app’s graspable.cloud address or its page on graspable.dev shows a large preview with its saved picture, title and public author name. Change Use in lists to choose the picture, or publish again to use a new project capture. Apps without a picture show a title card. Social networks may keep an older preview until they refresh their copy; existing posts do not always update. Private apps do not have public preview images. The gallery serves smaller copies of published project captures to fit your screen. Publishing a new capture gives it a new image address, so it can replace the previous cached picture. Nothing checks a picture beyond that it is a picture. The rules for apps cover their pages too, and every page has Report this app . On graspable.dev The same works on the website. Sign in at the top of graspable.dev with the account you use in the app (GitHub, Google, or a link by email), and every app's page has Like and Save , and every person's page has Follow . Your page ( graspable.dev/@yourhandle ) shows who you are, your apps with their pictures, how often they were viewed and liked, your followers, your public collections, and the month you joined. Under Account → Profile on the site ( graspable.dev/account/profile ) you set: - your handle and name; - a few words about you, up to 280 characters; - where you are, up to 60 characters; - up to five links, which must be https addresses. Each is shown with the name you have on that service (GitHub, X, YouTube, LinkedIn, Bluesky, Mastodon, Instagram and others) or by its host, and they are how people reach you: your email address is never shown; - what you build with, up to eight from a list of headsets and tools; - up to three of your public apps to show first on your page; - whether your page says you are open to work; - whether to show the picture of the GitHub or Google account you sign in with. It is off until you switch it on, and no other picture can be set. Under Account → Collections you choose which collections anyone can see, and remove a collection. - Views are counted per app, each time it is opened at its address, and shown on its page. - The picture on a card is the project's picture: the one you chose with Capture in the preview, or else the built app as a browser shows it when you publish. An app published before 0.12 gets its picture the next time you publish it. - The home page of graspable.dev shows the eight most viewed community apps. Nothing is placed there by hand. A private app is in none of these lists and cannot be liked or collected.","u":"/docs/publishing#community-views-likes-follows-and-collections"},{"k":"docs","t":"Publish and share","h":"Private apps","x":"On Builder and higher, an app can be private : it opens only with a link that holds a key. The key is shown once when you publish; make a new one under Settings → Published apps if a link got out. Private apps do not appear on your page and are not indexed.","u":"/docs/publishing#private-apps"},{"k":"docs","t":"Publish and share","h":"Continue from an app","x":"A public app shows a small Made with Graspable badge in the corner. It leads to the app's page on graspable.dev, where Continue from this app opens the app as a new project on the computer of whoever clicks it, with its editable source, ready for their next request. If you have not installed Graspable yet, download it. Every public app allows this unless you switch it off under Settings → Publishe","s":"A public app shows a small Made with Graspable badge in the corner. It leads to the app's page on graspable.dev, where Continue from this app opens the app as a new project on the computer of whoever clicks it, with its editable source, ready for their next request. If you have not installed Graspable yet, download it. Every public app allows this unless you switch it off under Settings → Published apps (Builder and higher); the badge goes away with it. The copy records which app it continued from.","u":"/docs/publishing#continue-from-an-app"},{"k":"docs","t":"Publish and share","h":"Native projects","x":"A Unity project or an Apple Vision Pro project runs on the device, not in a browser, so there is nothing to open at its address. Publishing one runs its check first, and what goes live is a page with the picture that check took (play mode in Unity, the visionOS Simulator for Vision Pro) and the project's source. People continue from it like from any app; to build it they need the same engine you d","s":"A Unity project or an Apple Vision Pro project runs on the device, not in a browser, so there is nothing to open at its address. Publishing one runs its check first, and what goes live is a page with the picture that check took (play mode in Unity, the visionOS Simulator for Vision Pro) and the project's source. People continue from it like from any app; to build it they need the same engine you do. See Native XR. The native examples under Made in Graspable are published this way. The headset app, as a download On Pro and higher, a Unity or Unreal Engine project can offer its headset app with the page. Choose Build for headset in the preview pane first, then Publish with Offer the headset app as a download ticked. The app's address, its page on graspable.dev and its example page then show a download button with the file's size and its SHA-256, so anyone can check that what they got is what you published. - The file is the Android app ( .apk ) from your last build. It is a test build signed with your computer's development key, not a store release, and the page says so. - A Meta Quest installs it only with developer mode on, from a computer: with Graspable, Meta Quest Developer Hub or adb install . For other Android headsets and glasses, follow their maker's instructions for installing an app file. - Publishing again without a new build does not send the file again. Settings → Published apps removes it. - One headset app per published app: up to 500 MB on Pro and 1 GB on Studio. A download counts as one visit. - An Apple Vision Pro app cannot be handed out as a file. Use TestFlight or the App Store and add the link, as below. Store pages When your app is in a store, add the address of its page there under Settings → Published apps → Store pages : the Meta Horizon Store, Google Play, the App Store, or a TestFlight invitation. The app's pages then show a button for each. This works on every plan and for every kind of app. Only addresses of those stores are accepted.","u":"/docs/publishing#native-projects"},{"k":"docs","t":"Publish and share","h":"Connectors in your app","x":"On Builder and higher, a published app can use one of the services you connected under Account → Connectors while it runs: read rows from your spreadsheet, add a line to it, post to your channel. The people who open the app sign in to nothing. The app calls a tool, and Graspable makes the call as you , with the values you pinned in place of whatever the app sent, so the app can only ever reach wha","s":"On Builder and higher, a published app can use one of the services you connected under Account → Connectors while it runs: read rows from your spreadsheet, add a line to it, post to your channel. The people who open the app sign in to nothing. The app calls a tool, and Graspable makes the call as you , with the values you pinned in place of whatever the app sent, so the app can only ever reach what you chose. 1. Publish the app, then open Account → Published apps → Edit its page on graspable.dev and go to Connectors . 2. Choose Grant a tool : a service you connected, one of its tools, and the values the app may not change: the spreadsheet, the folder, the channel. 3. In the project, the app calls the tool with the helper every template has: The call goes to the app's own address, connect/<service /<tool , so the app holds no key and no sign-in. In the preview on your computer the same grants work, through the Graspable app, once the project is published. Ask the agent for what the app should do with your data; it uses the helper and tells you which tool to grant. - Every tool that takes arguments is granted with at least one pinned value , because anyone who can reach the app can call the grant with any argument you did not pin. Tools that update, delete, share or change settings are not offered to apps. - Each call costs you 0.2 Credits , whether the service answers with a result or with an error. Each grant has its own limit of calls a day, 200 unless you change it, up to 2,000; the app also cannot make more than 30 calls a minute from one visitor or 600 a minute in all. - Switching a grant off stops the app's calls at once. Disconnecting the service stops them all. An app that rests for the month makes no calls. - An app's visitors can only ever call the tools you granted, with your pinned values. They cannot see your account, your other services or your connection. Pin everything a stranger should not choose.","u":"/docs/publishing#connectors-in-your-app"},{"k":"docs","t":"Publish and share","h":"AI in your app","x":"On Builder and higher, a published app can ask an AI model while it runs: answer a visitor's question, say what is in front of them from a picture of the scene or a camera frame, or hold a spoken conversation. The people who open the app need no key and no account, and no key is ever in the app. You pay for it from your Credits. 1. Publish the app, then open Account → Published apps → Edit its pag","s":"On Builder and higher, a published app can ask an AI model while it runs: answer a visitor's question, say what is in front of them from a picture of the scene or a camera frame, or hold a spoken conversation. The people who open the app need no key and no account, and no key is ever in the app. You pay for it from your Credits. 1. Publish the app, then open Account → Published apps → Edit its page on graspable.dev and go to AI in this app . 2. Switch it on and choose the model. You can add a short instruction that is sent before everything the app sends (\"You are the guide of this museum. Answer in two sentences.\"); the app cannot change it or leave it out. 3. In the project, the app asks with the helper every template has: The request goes to the app's own address, ai/chat , and Graspable asks the model for you. listen() and speak() use the browser's own speech recognition and voice, so a spoken conversation needs no other service; not every browser has them (some send the speech to their maker's service to recognise it), and the helper says when one is missing. In the preview on your computer the same AI answers, through the Graspable app, once the project is published. Ask the agent for what the app should do; it uses the helper and reminds you to switch AI on. - You pay by what each request costs , at the same rate as the Graspable agent on Credits: one Credit for every $0.10 the model provider charges. Each model's price, in Credits per million tokens (a token is about three quarters of a word), is shown where you choose it. Small requests add up over the day and are charged in tenths of a Credit. - Each app has a daily limit in Credits : 20 unless you change it, up to 500. When it is reached, or your Credits run out, the app gets a plain message instead of an answer until the next day. Answers are kept short: about 750 words unless the app asks for more, and about 1,500 at most. A request can carry up to four pictures. - One visitor can make up to 10 requests a minute, and the app 120 a minute in all. - Graspable keeps only counts of your app's requests (how many, how many pictures, tokens in and out, what they cost), never what was asked, a picture or an answer. The model provider receives the request to answer it. - Switching AI off stops the app's requests at once. An app that rests for the month makes none. - Anyone who can reach the app can ask through it. Choose the instruction and the daily limit with that in mind.","u":"/docs/publishing#ai-in-your-app"},{"k":"docs","t":"Publish and share","h":"How many, and how much","x":"Each plan comes with a number of published apps, private apps, visits per month, a size per app and, for native projects, a size for the headset app: Plan Apps Private Visits a month Size per app Headset app --- --- --- --- --- --- Explorer 2 — 2,000 60 MB — Builder 10 3 25,000 150 MB — Pro 30 10 100,000 250 MB 500 MB Studio 100 50 500,000 500 MB 1 GB An app that reaches its month's visits rests u","s":"Each plan comes with a number of published apps, private apps, visits per month, a size per app and, for native projects, a size for the headset app: Plan Apps Private Visits a month Size per app Headset app --- --- --- --- --- --- Explorer 2 — 2,000 60 MB — Builder 10 3 25,000 150 MB — Pro 30 10 100,000 250 MB 500 MB Studio 100 50 500,000 500 MB 1 GB An app that reaches its month's visits rests until the next month: its page says so instead of showing the app. Nothing is deleted. Enterprise plans have no limits; use the Contact sales button on the pricing page.","u":"/docs/publishing#how-many-and-how-much"},{"k":"docs","t":"Publish and share","h":"Rules, and reporting an app","x":"You are responsible for what you publish. Apps that deceive people, harm their devices, break the law or copy someone else's work are taken down. Anyone can report an app at graspable.dev/report; a person on the team reads every report.","u":"/docs/publishing#rules-and-reporting-an-app"},{"k":"docs","t":"Publish and share","h":"What is sent","x":"Publishing sends the built app and the project's source for people who continue from it. The source archive leaves out Git metadata, project memory, dependencies and generated build folders. Generated engine and film folders are omitted only at their project paths; source folders with matching names elsewhere are kept. It also leaves out files named .env , .env. (including examples), .npmrc , .net","s":"Publishing sends the built app and the project's source for people who continue from it. The source archive leaves out Git metadata, project memory, dependencies and generated build folders. Generated engine and film folders are omitted only at their project paths; source folders with matching names elsewhere are kept. It also leaves out files named .env , .env. (including examples), .npmrc , .netrc , .pypirc and .yarnrc.yml , at every folder depth, regardless of letter case. These are filename exclusions, not a scan for secrets: credentials embedded in code, assets or other files can still be published. Keep private credentials out of the code and assets you share. The exclusions apply to new publishes. Previously uploaded source archives are not rewritten.","u":"/docs/publishing#what-is-sent"},{"k":"docs","t":"Engines and sign-in","x":"Choose what powers the agent, sign in from inside the app, and learn which subscription or API key works where. Open Settings → AI for the agent . You choose one option and sign in right there. You need only one. Option What you sign in with Runs --- --- --- Graspable agent A ChatGPT or Claude subscription, an API key, a local model, or nothing: it can run on Credits Graspable's own agent Claude C","s":"Choose what powers the agent, sign in from inside the app, and learn which subscription or API key works where. Open Settings → AI for the agent . You choose one option and sign in right there. You need only one. Option What you sign in with Runs --- --- --- Graspable agent A ChatGPT or Claude subscription, an API key, a local model, or nothing: it can run on Credits Graspable's own agent Claude Code A Claude subscription (Pro or Max) Anthropic's Claude Code on your computer Codex CLI The ChatGPT login of the Codex CLI OpenAI's Codex on your computer. Shown only if the Codex CLI is installed Only the selected option shows its sign-in. All three edit the same project, show the same run panel and go through the same checks.","u":"/docs/engines-and-sign-in"},{"k":"docs","t":"Engines and sign-in","h":"Graspable agent","x":"Sign in with ChatGPT 1. Select Graspable agent . 2. Choose Sign in with ChatGPT . Your browser opens. 3. Finish signing in there. The Settings page updates by itself and shows \"Signed in with ChatGPT\". If the browser flow does not work, choose Not working? Use a code instead . The app shows a code to enter on the page that opened. This uses your ChatGPT plan through OpenAI's Codex sign-in. OpenAI ","s":"Sign in with ChatGPT 1. Select Graspable agent . 2. Choose Sign in with ChatGPT . Your browser opens. 3. Finish signing in there. The Settings page updates by itself and shows \"Signed in with ChatGPT\". If the browser flow does not work, choose Not working? Use a code instead . The app shows a code to enter on the page that opened. This uses your ChatGPT plan through OpenAI's Codex sign-in. OpenAI tolerates other apps using it but does not officially support them. If it stops working, switch to an API key. Sign in with Claude (experimental) 1. Select Graspable agent . 2. In Claude subscription , sign in with Claude. If needed, choose Install Claude Code first. 3. Finish in the browser, then choose Use Claude with the Graspable agent . An existing Claude sign-in works too. 4. Pick a model in the account row, or beside the prompt in a project. Graspable keeps its own tools, approvals, edit history and editing workflow. Model requests use the official, unmodified Claude connection on this computer. Graspable does not copy its sign-in credentials. The separate Claude Code option still runs Claude Code’s own agent. Some Claude models require reasoning; their account disables Off and asks you to choose Low , Medium or High if Off was saved earlier. Usage follows your Claude subscription and Agent SDK allowance. Extra usage depends on your Claude account settings; this does not promise unlimited or free requests. Available models also depend on that account. The subscription model picker does not show API token prices as subscription charges. The run’s dollar counter is marked est. : it uses API prices to enforce your budget limit and does not report a subscription bill. This option runs locally. Moving its project to the cloud blocks the run until you explicitly choose another account; Graspable does not silently replace it with an API key. API-key or custom-provider overrides in the local Claude environment are rejected for this option. On Credits, with no AI account of your own 1. Sign in to your Graspable account in Settings . 2. Under Graspable agent , choose Use the Graspable agent on Credits . The agent then runs on Graspable's AI account, and each request costs Credits by what it used: 10 Credits for every dollar the model costs. A new account has 6 Credits to try it. Settings shows the price of the chosen model, and you can choose any other model that can use tools there. Your own AI accounts keep working beside it and stay free. Requests go through Graspable to the model's provider. Graspable keeps only counts for them: tokens and Credits. Use an API key 1. Under Graspable agent , open Use an API key, or pay with crypto (or Add an API key, or pay with crypto ). 2. Pick a provider and a model. 3. Paste the key and choose Add . Get a key opens the provider's key page. The providers are Anthropic (Claude), OpenAI, OpenRouter, xAI, Groq, Mistral, Google Gemini, Vercel AI Gateway and Venice. Keys are stored in your computer's keychain and sent only to the agent running on your computer. An API key is billed per use by the provider. That is separate from, and on top of, any ChatGPT or Claude subscription. Venice Venice runs open and frontier models. The Graspable agent can use it with your Venice account, or with no account at all. With your Venice account Pick Venice in the form above and paste a key from venice.ai ( Settings → API ). The key uses whatever your account has, in the order Venice spends it: 1. Today's allowance , if you stake DIEM at Venice: one dollar a day for each DIEM staked. It renews at 00:00 UTC and what is unused does not carry over. 2. Plan credits , if your Venice plan includes them. 3. What you paid in , with a card or with crypto. Nothing has to be set up in Graspable for this: a key from an account that stakes or subscribes is spent that way by Venice itself. The account's row in Settings shows what is left of each, and Refresh reads it again. When nothing is left, a request is refused before it starts and says when the allowance renews. When you make the key at venice.ai, you can give it a spending limit per day. That is a good idea for a key you hand to any app. Paid with crypto, with no Venice account (experimental) This is new and has been tried less than the rest. Start with a small amount. Pick Venice, paid with crypto (experimental) and choose Create the agent's hot wallet . 1. Graspable makes a new wallet for the agent on this computer. It is yours: its key is kept in your keychain and you can show it at any time. It is a hot wallet (the key is on this computer, not in a hardware wallet), separate from your personal wallets, which Graspable never asks for. Keep in it only what you would let the agent spend. 2. Send USDC on the Base network to the address shown ( Copy address ), from an exchange or your own wallet. Other coins and other networks will not arrive. 3. Type an amount and choose Add to the Venice balance , then confirm. This sends that much USDC from the agent's wallet to Venice and cannot be taken back. The smallest amount is the one Venice names, $5 at the time of writing. 4. Requests are then charged against that balance, at Venice's price for the model you picked. The row shows what is in the wallet and what is left at Venice; Refresh reads both again. What to know: - Treat it like money in an app. Send what you plan to use, not savings. - Show the wallet's key shows the key so you can keep a copy or move what is left with another wallet app. Whoever has the key can spend what is in the wallet. - Remove deletes the key from this computer. Anything still in the wallet or in its Venice balance can then no longer be reached, so save the key first. - When the balance runs out, the run stops and says so. Add to the balance and send the request again. - The wallet stays on this computer. It cannot pay for a Cloud Agent; use a Venice API key there. - Graspable takes nothing from these payments and never holds your money. They go from the agent's wallet to Venice. Routstr, paid with bitcoin (experimental) Routstr is a network of servers (\"nodes\"), run by different people, that sell AI models for bitcoin. There is no account. This is new in Graspable and has been tried less than the rest: start with a small amount. 1. Pick Routstr, paid with bitcoin (experimental) in the form above. 2. Choose a node. Graspable lists nodes that have announced themselves and answer right now; Another node… takes an address you type. Then choose a model. Prices are shown in sats per million tokens, and differ from node to node. 3. Type an amount in sats and choose Get an invoice . Scan the QR code with a Lightning wallet, or copy the invoice into one, and pay it. 4. When the node sees the payment, the account appears by itself, with that balance. Requests are charged against it. The row then shows the balance. Get an invoice there adds to it. Withdraw what is left takes the balance out: to a Lightning address if you give one, otherwise as a Cashu token that you paste into a Cashu wallet. What to know: - A node is someone's server, and it holds your balance. Graspable checks that a listed node really announced itself and answers, not that the people behind it are honest. Pay what you plan to use, and withdraw the rest. - The invoice is remembered until the node says what became of it. If you pay and close the app, open Settings again and the account is set up then. Forget this invoice is only for an invoice you did not pay. - A Cashu token from a withdrawal is the money itself. Claim it in a wallet right away. While it is unclaimed, Show the last withdrawal again asks the node for it again. - Remove deletes the key of the balance from this computer. Withdraw first. - To change node , add another Routstr account with the other node: an account cannot be moved, because its balance is held by the node it was paid to. You can keep several and pick which one the agent uses. Confidential mode (experimental) Some nodes offer a confidential mode. With an ordinary node, the node passes your request on to a model provider, so it could read your requests and the answers, change what comes back, or bill for a dearer model than it used. In confidential mode the agent holds an encrypted connection with the provider itself, and the node only relays it: - the node cannot read your requests or the answers, and cannot change them, tool calls included; - the model and the token limit are proven to be the ones you asked for; - you pay for the usage the provider itself reported, on a receipt the node signs, and the agent checks all of it before a request counts as done. No special hardware is trusted: it rests on the provider's own certificate and on cryptography. The method is Juraj Bednar's, and Graspable uses his open-source client for it. Where a node offers it, a Confidential mode switch appears under the account's balance. Switching it on pins the node's signing key and the provider it names; if either changes later, requests stop until you switch it off and on again. What it costs you: - Speed and data. Every request is a session of its own: about 28 MB of traffic and around fifteen seconds before the first word. A long run makes many requests. - A larger balance in reserve. While a request runs, the node sets aside the cost of a full context for the model, and releases what was not used. - Answers up to 32,768 tokens , and requests up to 8 MB. What it does not hide: which model you use, when, and how long your requests and answers are. If a request cannot be made confidentially, it is not made at all. The method is new and has not had an outside review, so do not rely on it where a flaw would hurt you. - Graspable takes nothing from these payments and never holds your money. You pay the node from your own wallet. Confidential images. When your node offers image generation, select its Routstr account with Confidential mode switched on and ask the agent to list its image models. The list comes from the node's signed offer and includes the available quality tiers, their image-count limits and their prices in sats. Ask for an image, choose a tier, and approve the quoted price before generation starts. This is charged to your Routstr balance, separately from Graspable Credits and the agent's USDC wallet. Every generation needs your approval. Work without asking refuses these purchases. The node can see the image model, quality tier, count and request length. It cannot read your prompt or the generated image. The model provider sees both. Successful images use a fixed price and a signed receipt; there is no token-usage charge. Before you approve, the quote shows both the expected price and the maximum failed-session charge. A failed or interrupted request can still cost sats. The agent saves verified WebP images in a new folder under public/assets/ ; it never replaces an existing folder. The images become ordinary project assets that you and the agent can use and edit. No prompt or payment receipt is added to that public folder. Review your images before publishing a project, since assets under public/ are published with it. Generation takes a fresh encrypted session, with roughly 23 MB of setup traffic before the image downloads. Requests time out after eight minutes, and image sizes are bounded. If verification or saving fails, check the project files before approving a new attempt; a provider may already have charged for a completed generation. If the provider returns incomplete images or a response that disagrees with the approved request, those images are not saved; any verified charge is still recorded. There is no unencrypted retry. Confidential mode requires a node that supplies the current signed request rules and session authentication. Older or malformed offers are refused before a request starts. If a request stops with \"API key USD spend limit exceeded\" , this can be a limit on the node operator's upstream provider key even when your Routstr balance still has sats. Contact the node operator or choose another provider. Adding to your balance or creating another balance key may not resolve it. Use a local or custom model In the same form, pick Custom / local (OpenAI-compatible) . Enter the address and the model name. The key is optional. This works with any OpenAI-compatible chat endpoint, such as Ollama ( http://127.0.0.1:11434/v1 ), LM Studio, vLLM or a gateway. You can add more than one account. Choose Use this on another to switch.","u":"/docs/engines-and-sign-in#graspable-agent"},{"k":"docs","t":"Engines and sign-in","h":"Claude Code","x":"1. Select Claude Code . 2. If it is not installed, choose Install Claude Code . This takes about a minute. 3. Choose Sign in with Claude . Your browser opens. If the page shows a code, paste it into the field in Settings. Graspable runs your own, unmodified Claude Code. It never reads or copies your Claude sign-in. If your Claude Code is set up to use an API key or an Anthropic Console account, Se","s":"1. Select Claude Code . 2. If it is not installed, choose Install Claude Code . This takes about a minute. 3. Choose Sign in with Claude . Your browser opens. If the page shows a code, paste it into the field in Settings. Graspable runs your own, unmodified Claude Code. It never reads or copies your Claude sign-in. If your Claude Code is set up to use an API key or an Anthropic Console account, Settings shows a warning that usage is billed per use and not to your subscription.","u":"/docs/engines-and-sign-in#claude-code"},{"k":"docs","t":"Engines and sign-in","h":"Codex CLI","x":"This option appears only when the Codex CLI is installed on your computer. Choose Sign in with ChatGPT under it. You do not need it to use a ChatGPT subscription. The Graspable agent signs in with ChatGPT directly. Codex runs commands inside its own sandbox without asking you. See Approvals and safety.","u":"/docs/engines-and-sign-in#codex-cli"},{"k":"docs","t":"Engines and sign-in","h":"A different AI for one project","x":"Settings → AI for the agent sets the default for every project. To use something else in one project, open the project and choose it in the list above the prompt. The choice is kept for that project. Choose Default to follow Settings again.","u":"/docs/engines-and-sign-in#a-different-ai-for-one-project"},{"k":"docs","t":"Engines and sign-in","h":"Reasoning","x":"Reasoning , on the same page, sets how long the Graspable agent thinks before each step: Off , Low , Medium (the default) or High . More reasoning means better decisions on hard requests, and slower and costlier runs. The thinking is shown in the run panel as it streams. Claude Code and the Codex CLI decide this on their own; Claude Code does not show its reasoning text. In a cloud project only AP","s":"Reasoning , on the same page, sets how long the Graspable agent thinks before each step: Off , Low , Medium (the default) or High . More reasoning means better decisions on hard requests, and slower and costlier runs. The thinking is shown in the run panel as it streams. Claude Code and the Codex CLI decide this on their own; Claude Code does not show its reasoning text. In a cloud project only API key accounts can be chosen. If your default cannot run in the cloud, Graspable usually uses your first API key account and says so under the list. A Claude subscription choice stays blocked until you explicitly choose another account.","u":"/docs/engines-and-sign-in#reasoning"},{"k":"docs","t":"Engines and sign-in","h":"Which subscription works where","x":"You have Use --- --- ChatGPT subscription Graspable agent with Sign in with ChatGPT Claude subscription Graspable agent → Claude subscription (experimental), or Claude Code Anthropic API key Graspable agent Any other provider's API key Graspable agent A Venice account, with staked DIEM or a Venice plan Graspable agent with a Venice key Crypto, and no account anywhere Graspable agent with Venice, p","s":"You have Use --- --- ChatGPT subscription Graspable agent with Sign in with ChatGPT Claude subscription Graspable agent → Claude subscription (experimental), or Claude Code Anthropic API key Graspable agent Any other provider's API key Graspable agent A Venice account, with staked DIEM or a Venice plan Graspable agent with a Venice key Crypto, and no account anywhere Graspable agent with Venice, paid with crypto or Routstr, paid with bitcoin (both experimental) A model on your own computer Graspable agent with Custom / local Claude subscription requests use the official Claude connection. An Anthropic API key remains a separate, per-use option. Signing in with Google or GitHub Copilot to power the agent is not offered.","u":"/docs/engines-and-sign-in#which-subscription-works-where"},{"k":"docs","t":"Engines and sign-in","h":"Choosing a model","x":"Each option has a Model menu. Open it and type to search: every word you type must be in the model's name. Each row shows the model's name, how much text it can hold at once, what a million tokens cost to read and to write, and whether it reasons or reads pictures. Above the list you can narrow it to one maker, to models that reason or read pictures, and order it by newest, cheapest or largest. Wh","s":"Each option has a Model menu. Open it and type to search: every word you type must be in the model's name. Each row shows the model's name, how much text it can hold at once, what a million tokens cost to read and to write, and whether it reasons or reads pictures. Above the list you can narrow it to one maker, to models that reason or read pictures, and order it by newest, cheapest or largest. Where the list comes from: - OpenRouter : OpenRouter's own list, so a model is there the day OpenRouter adds it. - The Graspable agent on Credits : every OpenRouter model that can use tools and has a price. Graspable's own picks are marked and come first. Prices are shown in Credits. - Claude subscription under Graspable agent : the bundled Claude model catalog. A listed model is a candidate, not a guarantee that your account can use it. - Other API keys and the ChatGPT sign-in : a public catalog of each provider's models. - Venice and Routstr : Venice's list, or the node's. - Claude Code and the Codex CLI : the tool installed on your computer. Default means whatever that tool is set to. For one project Above the prompt in a project, next to AI , the same menu changes the model for that project alone. Settings keeps the default for the others.","u":"/docs/engines-and-sign-in#choosing-a-model"},{"k":"docs","t":"Engines and sign-in","h":"Keeping the lists current","x":"Graspable keeps the lists on your computer and shows them at once. Each time it starts, it asks every source again in the background, so a model released yesterday is in the menu today without a new version of Graspable. Refresh model list at the top of the section asks again right away. At the same time Graspable checks whether Claude Code and the Codex CLI on your computer are their newest versi","s":"Graspable keeps the lists on your computer and shows them at once. Each time it starts, it asks every source again in the background, so a model released yesterday is in the menu today without a new version of Graspable. Refresh model list at the top of the section asks again right away. At the same time Graspable checks whether Claude Code and the Codex CLI on your computer are their newest versions, because a newer one can bring new models. When one is behind, its option says so and offers Update , which runs the tool's own updater. Nothing is updated without that click.","u":"/docs/engines-and-sign-in#keeping-the-lists-current"},{"k":"docs","t":"Engines and sign-in","h":"Signing out","x":"- Graspable agent: Sign out next to the ChatGPT row, or Remove next to a key. - Claude subscription under Graspable agent: Remove removes the choice from Graspable. To end the shared Claude sign-in too, use Sign out under Claude Code . - Claude Code and Codex CLI: Sign out under the option. A project that runs in the cloud needs the Graspable agent with an API key. See Cloud agents.","u":"/docs/engines-and-sign-in#signing-out"},{"k":"docs","t":"Approvals and safety","x":"What the agent may do on its own, what it asks you first, how System One and run budgets work, and what the agent cannot touch. The agent works in one place: your project folder. It reads and writes files there and can run commands there. This page explains what it does freely and what waits for you.","u":"/docs/approvals-and-safety"},{"k":"docs","t":"Approvals and safety","h":"What is asked","x":"Each thing the agent does has a risk level. Risk Examples In the app --- --- --- Read Reading and searching files Runs without asking Write Creating and editing files in the project Runs without asking Command Running a shell command Asks first External Fetching a web page, searching the web Asks first The app runs with these rules. There is no setting in the app to change them. The agent software","s":"Each thing the agent does has a risk level. Risk Examples In the app --- --- --- Read Reading and searching files Runs without asking Write Creating and editing files in the project Runs without asking Command Running a shell command Asks first External Fetching a web page, searching the web Asks first The app runs with these rules. There is no setting in the app to change them. The agent software also has a stricter mode in which file writes ask too; the app does not offer it yet. With Claude Code , reading and editing are pre-approved only inside the project folder, and the project's build commands are pre-approved. Anything else, including a write outside the project, shows the approval card. With the Codex CLI , there are no approval cards. Codex runs commands inside its own sandbox, limited to the project folder, without asking. Settings shows a warning about this under the option.","u":"/docs/approvals-and-safety#what-is-asked"},{"k":"docs","t":"Approvals and safety","h":"The approval card","x":"Waiting for your answer pauses the agent's inactivity timer. Approval requests still expire after five minutes and count as declined. If a run ends or is cancelled, its unanswered cards close; they cannot approve a purchase or action afterward. When something needs your approval, the run pauses and an amber card appears with: - the name of the tool and its risk, - the command, or for a file change","s":"Waiting for your answer pauses the agent's inactivity timer. Approval requests still expire after five minutes and count as declined. If a run ends or is cancelled, its unanswered cards close; they cannot approve a purchase or action afterward. When something needs your approval, the run pauses and an amber card appears with: - the name of the tool and its risk, - the command, or for a file change the exact difference it would make to the file as it is now, - a warning if an edit cannot apply, - an \"estimate\" line, when System One has judged the action. Choose Allow or Deny . With the Graspable agent, a card left unanswered for five minutes counts as a denial. After a denial the agent is told and carries on another way, or stops. If an approval expires while you are answering it, the old card closes. If the answer cannot reach the agent, the card stays open with an error so you can try again.","u":"/docs/approvals-and-safety#the-approval-card"},{"k":"docs","t":"Approvals and safety","h":"System One, in plain words","x":"System One is a small, fast model that answers narrow questions such as \"is this command harmless?\" in about a third of a second. It never writes code or text. When you are signed in to your Graspable account, Graspable's own System One model is used. You do not have to set anything up. Signed out, only Graspable's fixed rules decide. What it does in a run: It decides What happens --- --- Is this ","s":"System One is a small, fast model that answers narrow questions such as \"is this command harmless?\" in about a third of a second. It never writes code or text. When you are signed in to your Graspable account, Graspable's own System One model is used. You do not have to set anything up. Signed out, only Graspable's fixed rules decide. What it does in a run: It decides What happens --- --- Is this command harmless? A harmless command is approved for you, but only when the fixed rules already find it clean Is this saved note about your request? Notes that are clearly about something else are left out Is this lesson worth remembering? Version numbers and one-off details are not saved to memory Is this a question, a small edit or a build? After a question, the \"what did we learn\" step is skipped Is the agent repeating the same failed fix? The run stops instead of spending more of your AI account Can a code change fix this failing check? Repair turns that cannot help are not started Does the final message claim more than the run checked? You see a note under the message Does this command output or package file try to give the agent instructions? A warning is put in front of it, and the agent is told to treat it as data Which parts of a long command output matter? The rest is left out before the agent reads it After a project moved: which parts of the earlier conversation does your request need? Only those are handed to the agent A scene differs from its XR recording: is that what you asked for? You see a hint; you still decide whether to accept the new scene What it cannot do: - deny anything. A doubtful action always goes to you. - approve changes to files Graspable manages (listed below). - accept a changed XR scene for you. The modes, under Settings → System One (fast decisions) : Mode Meaning --- --- Assist The decisions above are acted on Shadow Judge and log only. Nothing is decided for you Off No System One calls Every judgement appears in the run with a ⚡ mark. What it costs System One is included with a Graspable account. Each day a generous amount is free: about 400 decisions, more than a normal day of building uses. Beyond that, paid plans use Credits: 0.1 Credit for about every 130 further decisions. On the free plan it pauses until the next day, and the fixed rules decide meanwhile. Your own model You can add your own model instead: Jev (TypeSafe), any OpenAI-compatible endpoint, or Anthropic's Claude Haiku. Yours is then asked first and Graspable's is the fallback. Test shows what a model would do with a harmless read and with a command that would leak a secret.","u":"/docs/approvals-and-safety#system-one-in-plain-words"},{"k":"docs","t":"Approvals and safety","h":"Helpers","x":"The Graspable agent can hand a self-contained part of a request to a helper: a second conversation with the same model and the same tools, working in the same project. The helper knows only what the agent wrote in its task, and it cannot start helpers of its own. Its tool calls show in the run panel like the agent's, its commands and file changes go through the same approval cards, and what it spe","s":"The Graspable agent can hand a self-contained part of a request to a helper: a second conversation with the same model and the same tools, working in the same project. The helper knows only what the agent wrote in its task, and it cannot start helpers of its own. Its tool calls show in the run panel like the agent's, its commands and file changes go through the same approval cards, and what it spends counts against the run budget. Helpers the agent starts in the same step work at the same time, each on its own files; its other tools run one at a time. The agent uses at most three helpers in a request, and most requests need none.","u":"/docs/approvals-and-safety#helpers"},{"k":"docs","t":"Approvals and safety","h":"Run budget","x":"In Settings → Run budget you can set hard caps per run: max $ and max tokens . Leave a field empty for no cap. When the reported usage crosses a cap, the run is stopped. Cost is known only for providers that report it. Tokens are always counted. With Claude Code and the Codex CLI the check happens between turns, so a run can pass the cap slightly before it stops. Settings → Usage shows totals for ","s":"In Settings → Run budget you can set hard caps per run: max $ and max tokens . Leave a field empty for no cap. When the reported usage crosses a cap, the run is stopped. Cost is known only for providers that report it. Tokens are always counted. With Claude Code and the Codex CLI the check happens between turns, so a run can pass the cap slightly before it stops. Settings → Usage shows totals for every run on your computer, whichever way the AI was paid for, and saves them as a file. Where AI was paid for with crypto (Venice from the agent's wallet, a Routstr node), a second list shows each request and each filling of the balance with the coin amount, the dollars it stood for and the time. Graspable also stops a run that is going nowhere, such as the same action repeated again and again. The transcript shows a ⚠ line saying why.","u":"/docs/approvals-and-safety#run-budget"},{"k":"docs","t":"Approvals and safety","h":"What the agent cannot touch","x":"- Files outside the project. File tools accept only paths inside the project folder. Paths that try to leave it, including through links, are refused. - .graspable/ in the project. It holds memory and the run journal. It is hidden from the file list and cannot be written by the agent's file tools. - tests/xr/ , xr-dev.js and graspable-source.js . The agent's file tools refuse to write them, whiche","s":"- Files outside the project. File tools accept only paths inside the project folder. Paths that try to leave it, including through links, are refused. - .graspable/ in the project. It holds memory and the run journal. It is hidden from the file list and cannot be written by the agent's file tools. - tests/xr/ , xr-dev.js and graspable-source.js . The agent's file tools refuse to write them, whichever engine runs; a recorded session is written only through verify app after you confirm it. - Earlier conversations, captures and memory files. The agent cannot read them back through its tools; it gets memory only the way described under Memory. - Files that run code ( package.json , the Vite config, npm's own config) are a different matter: a change to one of them asks you like a command does, because the next build, preview or install would run it. Version 0.19.1 also checks the destination of linked files before the Graspable agent changes build configuration. A different link name cannot turn that change into an ordinary file edit. Linked source files keep their usual approval behavior. Text the agent reads from outside the conversation (command output, a page or message a connected service returned, a plugin's answer, every file of a project you continued from) is marked as outside text when it reads like instructions to the agent, and the agent is told to treat it as data. The agent also refuses to remember such text. A plugin you add from the web does not decide which of its tools may run without asking: every one asks. - Your sign-ins. Graspable never reads or copies the sign-in of Claude Code or the Codex CLI. A shell command is a different matter: once you allow a command, it runs with your user's permissions. Read the card before you choose Allow .","u":"/docs/approvals-and-safety#what-the-agent-cannot-touch"},{"k":"docs","t":"Approvals and safety","h":"If something goes wrong","x":"Every file change in a run is recorded. Undo run puts the files back. See Your first project.","u":"/docs/approvals-and-safety#if-something-goes-wrong"},{"k":"docs","t":"Memory","x":"What the agent remembers about you and your projects, where it is stored, and how to pin, edit or forget an item. Memory lets the agent keep what it learned from one run to the next: your preferences, decisions made in a project, things that failed and why. Memory is files on your computer. It is not sent to Graspable.","u":"/docs/memory"},{"k":"docs","t":"Memory","h":"Two kinds","x":"Kind About Stored in Managed in --- --- --- --- Your memory You, across all projects ~/.graspable/memory Settings → Memory Project memory One project .graspable/memory inside the project folder Project memory , under the file list Each item is one Markdown file. A file named MEMORY.md in the same folder is a generated index for people to read.","u":"/docs/memory#two-kinds"},{"k":"docs","t":"Memory","h":"What the agent knows about your earlier requests","x":"Memory is for what should last. Apart from it, the agent also needs to know what you asked a moment ago. - Every request starts with a short account of the last few. For each of the last five requests: what you asked, which files changed and how it ended. It is made on your computer from the project's run log, without asking a model. - A quick follow-up goes on with the same conversation. With the","s":"Memory is for what should last. Apart from it, the agent also needs to know what you asked a moment ago. - Every request starts with a short account of the last few. For each of the last five requests: what you asked, which files changed and how it ended. It is made on your computer from the project's run log, without asking a model. - A quick follow-up goes on with the same conversation. With the Graspable agent, when you ask again within about four minutes and nothing in the project changed in between, the agent continues where it stopped, so \"make it bigger\" needs no looking around first. After a longer pause, an undo, an edit of your own or a change of model, it starts a new conversation with the short account above. That is deliberate: carrying a long conversation along costs more than starting anew once the AI provider no longer has it at hand. - Where it is kept. The conversation is kept step by step in the project's .graspable folder, on your computer or on the project's cloud agent. It is not published with an app, not part of the project's history and not sent to Graspable. Because every step is kept as it happens, a run that the app interrupted (an update, a crash) continues from its last step when you open the project again. From version 0.19.1, Graspable releases conversation database resources when a run ends, without waiting for automatic memory cleanup. Saved conversations remain available for follow-up requests.","u":"/docs/memory#what-the-agent-knows-about-your-earlier-requests"},{"k":"docs","t":"Memory","h":"How items get there","x":"You add them. - In Settings → Memory , pick a kind, type a title and what to remember, and choose Remember . - In a project, type into Remember about this project… and press Enter. The agent adds them. During a run the agent can save something it learned. After a run that changed files, it is also asked for a few lasting lessons, which are saved as project memory. The transcript shows these as \"🧠","s":"You add them. - In Settings → Memory , pick a kind, type a title and what to remember, and choose Remember . - In a project, type into Remember about this project… and press Enter. The agent adds them. During a run the agent can save something it learned. After a run that changed files, it is also asked for a few lasting lessons, which are saved as project memory. The transcript shows these as \"🧠 remembered\". Anything that looks like a secret, such as a key or password, is refused.","u":"/docs/memory#how-items-get-there"},{"k":"docs","t":"Memory","h":"How the agent uses memory","x":"The agent does not get your whole memory every time. It gets two things. - The most important items, always. These are the items you pinned, then what you said yourself, then conventions, preferences and decisions, then the items used most often. This part has a size limit. - A few items that match the request. For each request, Graspable looks for items that relate to what you asked and adds at m","s":"The agent does not get your whole memory every time. It gets two things. - The most important items, always. These are the items you pinned, then what you said yourself, then conventions, preferences and decisions, then the items used most often. This part has a size limit. - A few items that match the request. For each request, Graspable looks for items that relate to what you asked and adds at most six. When nothing matches, nothing is added. The same happens again on a repair turn, where past failures and lessons count for more.","u":"/docs/memory#how-the-agent-uses-memory"},{"k":"docs","t":"Memory","h":"\"Recalled\" in the transcript","x":"When items are added for a request, the transcript shows a line such as: Hover over the line to see each item's kind and where it lives. On a repair turn the line reads \"recalled for the fix\". To see the exact text the agent received, open Context at the top of the run panel and look for the recalled memory piece.","u":"/docs/memory#quot-recalled-quot-in-the-transcript"},{"k":"docs","t":"Memory","h":"Open, pin, edit, forget","x":"Click an item to read it. Under it you see where it came from (\"You said this\" or \"Learned by the agent\"), when, and how often it was used. Move the pointer over an item to see its buttons. - Pin — the agent always gets this item. Click again to unpin. - Edit — change the title or text. An edited item counts as something you said. - Forget — the agent no longer gets it. The file stays in the memor","s":"Click an item to read it. Under it you see where it came from (\"You said this\" or \"Learned by the agent\"), when, and how often it was used. Move the pointer over an item to see its buttons. - Pin — the agent always gets this item. Click again to unpin. - Edit — change the title or text. An edited item counts as something you said. - Forget — the agent no longer gets it. The file stays in the memory folder, marked as archived. In Settings → Memory , Tidy up merges items that say the same thing.","u":"/docs/memory#open-pin-edit-forget"},{"k":"docs","t":"Memory","h":"When items change","x":"- Saving something that is already known does nothing. - A newer item about the same subject replaces the older one. The older file is kept and marked as replaced. - What the agent infers never replaces what you said. - Nothing is deleted for being old. An item marked ⚠ was learned with an older version of Graspable. Check that it still holds, then edit or forget it.","u":"/docs/memory#when-items-change"},{"k":"docs","t":"Memory","h":"Limits","x":"- Matching is by words. A request that shares no word with an item will not recall it. Pin the items that matter most. - With the Codex CLI, the agent receives memory but cannot save new items during a run. - Memory helps the agent stay consistent. It is not a guarantee that the agent follows every item.","u":"/docs/memory#limits"},{"k":"docs","t":"Memory","h":"In cloud projects","x":"When you move a project to the cloud, its memory and your memory travel with it and come back when you move it back. See Cloud agents.","u":"/docs/memory#in-cloud-projects"},{"k":"docs","t":"Voice","x":"Speak your requests to the agent, or talk with it and hear its answers. Speech goes to ElevenLabs, on Credits or with your own ElevenLabs key. You can speak to the agent in three ways: say a request instead of typing it, hold a conversation in which the agent answers aloud, or talk with it from inside VR or AR.","u":"/docs/voice"},{"k":"docs","t":"Voice","h":"Before you start","x":"Voice needs one of two things, chosen under Settings, Voice : - your Graspable account, signed in. Voice then runs on Credits, with free minutes every day; or - your own ElevenLabs API key. Who turns speech into text says what each costs.","u":"/docs/voice#before-you-start"},{"k":"docs","t":"Voice","h":"Speaking a request","x":"Press the microphone beside the prompt, say what you want, and press it again (or Enter). What you said appears in the prompt, where you can change it before you send it. Escape drops the recording. Graspable listens only between the two presses. The first time, your computer asks whether Graspable may use the microphone. One recording holds a few minutes of speech; for a longer request, speak it ","s":"Press the microphone beside the prompt, say what you want, and press it again (or Enter). What you said appears in the prompt, where you can change it before you send it. Escape drops the recording. Graspable listens only between the two presses. The first time, your computer asks whether Graspable may use the microphone. One recording holds a few minutes of speech; for a longer request, speak it in parts.","u":"/docs/voice#speaking-a-request"},{"k":"docs","t":"Voice","h":"Talking with the agent","x":"The button with the sound waves, under the microphone, starts a conversation. The microphone then stays open: - Say what you want. When you stop speaking, what you said is sent: as a new request, or, while the agent is working, as a note that steers it. - The agent answers in a few short sentences, and they are read aloud. A request you speak asks the agent to answer this way. Code and tables are ","s":"The button with the sound waves, under the microphone, starts a conversation. The microphone then stays open: - Say what you want. When you stop speaking, what you said is sent: as a new request, or, while the agent is working, as a note that steers it. - The agent answers in a few short sentences, and they are read aloud. A request you speak asks the agent to answer this way. Code and tables are left out of what is read; they stay in the run panel. - Speak while it reads and the reading stops, so you can interrupt it. - When the agent needs your OK for a step, it says so. Approve or decline on the screen. Press the button again, or the cross beside \"Listening\", to end the conversation and close the microphone. To keep the answers silent, turn off Read the agent's answers aloud under Settings, Voice . Every agent message also has a Read aloud button beside Copy.","u":"/docs/voice#talking-with-the-agent"},{"k":"docs","t":"Voice","h":"Talking in the headset","x":"You can also talk with the agent while you are in VR or AR, in a web project. - In the preview on this computer (the emulated headset), a Talk to the agent button sits low in front of you. Pressing it starts the same conversation as the button beside the prompt, with this computer's microphone. - On a real headset , open Headset , share the preview on Wi-Fi, and choose Talk in the headset . Scan t","s":"You can also talk with the agent while you are in VR or AR, in a web project. - In the preview on this computer (the emulated headset), a Talk to the agent button sits low in front of you. Pressing it starts the same conversation as the button beside the prompt, with this computer's microphone. - On a real headset , open Headset , share the preview on Wi-Fi, and choose Talk in the headset . Scan the new code with the headset. In VR or AR, press Talk to the agent and speak. The headset listens with its own microphone, and the agent's answers play in the headset. The conversation also shows in the app on your computer, where you approve steps. The headset sends its recordings to Graspable on your computer, which turns them into text the same way as above. The headset never holds your ElevenLabs key or your Graspable sign-in. Choose Stop talking in the headset , or stop sharing, to end it.","u":"/docs/voice#talking-in-the-headset"},{"k":"docs","t":"Voice","h":"Who turns speech into text","x":"ElevenLabs turns speech into text, and the agent's answers into speech. Choose how under Settings, Voice : - Graspable, on Credits. Needs your Graspable account. About ten minutes of speech every day are free. Beyond them, voice uses Credits by what it cost: about a third of a Credit for ten more minutes of speaking, and about a Credit for a page of answers read aloud. - ElevenLabs, with your own ","s":"ElevenLabs turns speech into text, and the agent's answers into speech. Choose how under Settings, Voice : - Graspable, on Credits. Needs your Graspable account. About ten minutes of speech every day are free. Beyond them, voice uses Credits by what it cost: about a third of a Credit for ten more minutes of speaking, and about a Credit for a page of answers read aloud. - ElevenLabs, with your own key. ElevenLabs bills your own account. The key stays in your computer's keychain and is sent only to ElevenLabs. Either way, the recording goes from the app to the agent on your computer, and from there to ElevenLabs (directly with your own key, or through Graspable). Graspable keeps no recording and no text of what you said, only how many seconds were turned into text and how many characters were read aloud that day.","u":"/docs/voice#who-turns-speech-into-text"},{"k":"docs","t":"Voice","h":"Language","x":"Speech to text finds the language by itself. If it guesses wrong, for example with short requests, choose the language you speak under Settings, Voice .","u":"/docs/voice#language"},{"k":"docs","t":"Voice","h":"The agent's voice","x":"Under Settings, Voice you also choose the voice the agent answers with, and Hear it plays a sample. On Credits the choice is ElevenLabs' own voices. With your own key it is every voice in your ElevenLabs account, including voices you made there.","u":"/docs/voice#the-agent-39-s-voice"},{"k":"docs","t":"Voice","h":"If something does not work","x":"The app says what went wrong beside the prompt, for example that it may not use the microphone or that none was found. Troubleshooting lists each message and what to do.","u":"/docs/voice#if-something-does-not-work"},{"k":"api","t":"Use Graspable with your AI agent","x":"For people and for AI agents. How an agent such as Claude Code, Codex, OpenCode or Amp installs, sets up and uses Graspable to build XR projects. Your coding agent can use Graspable to create a project, edit its files, check the scene and open a preview. It can do the editing itself through MCP, using the AI account it already has. You can also delegate a request to Graspable's agent. Web project ","s":"For people and for AI agents. How an agent such as Claude Code, Codex, OpenCode or Amp installs, sets up and uses Graspable to build XR projects. Your coding agent can use Graspable to create a project, edit its files, check the scene and open a preview. It can do the editing itself through MCP, using the AI account it already has. You can also delegate a request to Graspable's agent. Web project checks build the project and load it in a browser; native projects use checks for their engine. This page is written so that an agent can follow it too.","u":"/docs/use-with-ai-agents"},{"k":"api","t":"Use Graspable with your AI agent","h":"What Graspable is good for","x":"Give Graspable a task when the goal is a 3D, AR or VR experience for the web: a WebXR scene, game, product viewer, training or measuring tool that runs in a browser and on headsets such as Meta Quest and Apple Vision Pro. These WebXR scenes use React Three Fiber with @react-three/xr , built with Vite. AI glasses use the separate paths below. Native templates use Unity, Unreal Engine or Xcode and n","s":"Give Graspable a task when the goal is a 3D, AR or VR experience for the web: a WebXR scene, game, product viewer, training or measuring tool that runs in a browser and on headsets such as Meta Quest and Apple Vision Pro. These WebXR scenes use React Three Fiber with @react-three/xr , built with Vite. AI glasses use the separate paths below. Native templates use Unity, Unreal Engine or Xcode and need the corresponding tools and plan access. Create premium or native projects in the desktop app first, then select them through MCP with list projects . Immersive features depend on the browser and device. The Tabletop AR web template uses its tracking mat on iPhone; use native phone AR for bare-table placement. See templates and Native XR.","u":"/docs/use-with-ai-agents#what-graspable-is-good-for"},{"k":"api","t":"Use Graspable with your AI agent","h":"Build an AI glasses app with your agent","x":"Graspable also has templates for everyday AI glasses. Choose the target before asking the agent to create a project: - Meta Ray-Ban Display: Display glasses web app is a small web interface for timers, checklists and instructions. It has no camera or microphone access. - Ray-Ban Meta or Oakley Meta: Meta glasses app is an Expo phone companion using Meta's Device Access Toolkit. Glasses field notes","s":"Graspable also has templates for everyday AI glasses. Choose the target before asking the agent to create a project: - Meta Ray-Ban Display: Display glasses web app is a small web interface for timers, checklists and instructions. It has no camera or microphone access. - Ray-Ban Meta or Oakley Meta: Meta glasses app is an Expo phone companion using Meta's Device Access Toolkit. Glasses field notes starts with a photo journal and spoken notes. Experimental features can be unavailable on a particular device or account. These templates need Builder or higher. Create the project in Graspable, then have your connected agent find it with list projects , inspect its files and make the changes. The catalog reports these as kind: web because they have browser previews; the phone companions still need a native phone build to connect to real glasses. Keep the existing Expo or web runtime when editing. The browser preview helps check screens and simulated input. For a companion app, use Run on my phone in Graspable and test the actual glasses. For a Display web app, publish only when you are ready to share it, then test installation and controls on the glasses. A passing browser check does not establish device support or permission to distribute through Meta. Follow the glasses guide for tools, account requirements and supported capabilities.","u":"/docs/use-with-ai-agents#build-an-ai-glasses-app-with-your-agent"},{"k":"api","t":"Use Graspable with your AI agent","h":"Set-up","x":"1. Install grasp. The Graspable app can install it from Settings → Plugins and connectors → Use Graspable from other tools . The app is available for Windows, macOS and Linux. Agents preparing an unattended environment can use the platform installers documented in CLI. The app is optional for command-line use. 2. Check the connection. grasp projects works whether or not the app is open. 3. Connect","s":"1. Install grasp. The Graspable app can install it from Settings → Plugins and connectors → Use Graspable from other tools . The app is available for Windows, macOS and Linux. Agents preparing an unattended environment can use the platform installers documented in CLI. The app is optional for command-line use. 2. Check the connection. grasp projects works whether or not the app is open. 3. Connect your agent in one of two ways: - MCP server , best for agents that support MCP: claude mcp add graspable -- grasp mcp , codex mcp add graspable -- grasp mcp , or the JSON in MCP. - The command line , for any agent that can run commands: see CLI. Creating a free React XR project, editing its files and running local checks do not require another AI sign-in or spend Graspable Credits. Your external agent's own usage terms still apply. Premium templates, hosted services and paid asset generation have their own requirements. If you want Graspable's agent to generate or change the scene, configure its AI account through the app's Settings → AI for the agent or the interactive grasp login flow. That account's allowance or balance pays for delegated runs. A person completes any browser sign-in; an agent should never ask for sign-in tokens in chat.","u":"/docs/use-with-ai-agents#set-up"},{"k":"api","t":"Use Graspable with your AI agent","h":"Instructions for the agent","x":"Install the Graspable skill Ask your agent to install the Graspable skill for the current project. It covers project creation, file edits, run progress, checks, previews and device requirements. Installing the skill adds instructions; it does not install Graspable, sign in or grant approvals. For agents using the skills client, the project-scoped command is: The client asks which agent should rece","s":"Install the Graspable skill Ask your agent to install the Graspable skill for the current project. It covers project creation, file edits, run progress, checks, previews and device requirements. Installing the skill adds instructions; it does not install Graspable, sign in or grant approvals. For agents using the skills client, the project-scoped command is: The client asks which agent should receive it. Agents without that client can read the skill directly or use the short instructions below. The discovery index includes a SHA-256 digest of the skill file. Short instructions You can paste this into your agent's instructions file ( CLAUDE.md , AGENTS.md or similar):","u":"/docs/use-with-ai-agents#instructions-for-the-agent"},{"k":"api","t":"Use Graspable with your AI agent","h":"With the MCP server","x":"For direct editing, use create project → wait for installation → list files / read file → write file → check project → start preview . Read each file before changing it. A failed check needs a correction and another check. The preview address works on that computer; sharing a public link is a separate publishing step. Native projects use their engine checks and the desktop preview instead of start","s":"For direct editing, use create project → wait for installation → list files / read file → write file → check project → start preview . Read each file before changing it. A failed check needs a correction and another check. The preview address works on that computer; sharing a public link is a separate publishing step. Native projects use their engine checks and the desktop preview instead of start preview ; see the native MCP workflow. With Graspable 0.19.0 or later, check project can return an image alongside its report. Request includeImage: true when your agent needs to inspect the result. It uses the available check screenshot; it does not capture a new headset session. See check screenshots for cached checks, missing images and size limits. For a complete editable example, follow the WebXR inspection exhibit walkthrough. It includes the source component, a real capture and desktop and emulated-controller checks. For delegated generation, use start run → get run until it finishes, then open the web or native preview as above. Follow the same run after a disconnect. Pass the returned since value on the next poll to receive only new events. get run waits up to ten minutes per call. The MCP guide covers both workflows and their tools.","u":"/docs/use-with-ai-agents#with-the-mcp-server"},{"k":"api","t":"Use Graspable with your AI agent","h":"Long-running and background work","x":"Runs keep going when the caller disconnects. Start one with grasp run … --no-wait , do other work, and come back with grasp logs <run --follow . To keep a project running while the computer sleeps, move it to a Cloud Agent in the app.","u":"/docs/use-with-ai-agents#long-running-and-background-work"},{"k":"api","t":"Use Graspable with your AI agent","h":"What your agent may do","x":"The MCP server uses run-level access: it can create projects, edit files, start runs and answer approval requests. It cannot change plugins, sign-ins or settings, and it cannot delete projects. The CLI also has account and settings commands; use those only when the person has requested that setup. For a script with narrower rights, create a read token: API. Your agent cannot spend your money throu","s":"The MCP server uses run-level access: it can create projects, edit files, start runs and answer approval requests. It cannot change plugins, sign-ins or settings, and it cannot delete projects. The CLI also has account and settings commands; use those only when the person has requested that setup. For a script with narrower rights, create a read token: API. Your agent cannot spend your money through Graspable. Agent spending, which lets Graspable's agent pay outside services from a wallet of its own, applies only to requests started in the Graspable app. Runs started through grasp or MCP never get the tools to pay, because that wallet's key stays in the app's keychain. Delegated runs still use the AI account and budget described above.","u":"/docs/use-with-ai-agents#what-your-agent-may-do"},{"k":"api","t":"Use Graspable with your AI agent","h":"Read these pages as Markdown","x":"Every documentation page has a Markdown version at the same address with .md at the end, and the whole documentation is listed at graspable.dev/llms.txt. Links between the Markdown references keep you in plain text for docs, articles, templates, comparisons and pricing. Prices, annual billing and plan limits are available at pricing.md. agents.json lists the public interfaces and templates. These ","s":"Every documentation page has a Markdown version at the same address with .md at the end, and the whole documentation is listed at graspable.dev/llms.txt. Links between the Markdown references keep you in plain text for docs, articles, templates, comparisons and pricing. Prices, annual billing and plan limits are available at pricing.md. agents.json lists the public interfaces and templates. These documents describe capabilities; they do not grant permission to sign in, purchase services or answer approvals for a person.","u":"/docs/use-with-ai-agents#read-these-pages-as-markdown"},{"k":"api","t":"Use Graspable with your AI agent","h":"For agents browsing the site","x":"On browsers offering experimental WebMCP, public pages expose two read-only tools: search graspable finds guides, templates, skills, comparisons and pricing; read graspable reference reads a result as Markdown. Search by topic or ask a question, optionally filter by category, and follow nextOffset to continue a long page. Search ignores common question words such as “how” and “can”; the remaining ","s":"On browsers offering experimental WebMCP, public pages expose two read-only tools: search graspable finds guides, templates, skills, comparisons and pricing; read graspable reference reads a result as Markdown. Search by topic or ask a question, optionally filter by category, and follow nextOffset to continue a long page. Search ignores common question words such as “how” and “can”; the remaining words must match. Documentation sections and individual pricing plans are searchable beyond their opening paragraph, with a short excerpt around the match. Plan results link to the matching plan on the pricing page. Comparison results can link to a specific row; the result heading identifies whose feature or limit is described. Follow the full reference for source links and the date its facts were checked. If a question returns no result, try the main topic in fewer words. Search uses the same public index as the site's search button. Queries stay in the browser. The tools fetch public pages without account cookies. Browser support varies; the links and Markdown pages work without WebMCP. Creating and editing a project uses the local CLI or MCP connection described above. Articles are searchable by section, including instructions beyond their opening description. A result links to the matching heading; reading it returns the article's Markdown, including its cover image and update date when provided. For example, search for Rhubarb mouth-cue to find the speech-import walkthrough, or inputsourceschange for the guide to changing XR input sources.","u":"/docs/use-with-ai-agents#for-agents-browsing-the-site"},{"k":"api","t":"Plugins and connectors","x":"Connect your accounts of other services, and give the agent tools from other programs. How to add a plugin, how to write one, and the manifest format. A plugin gives the agent more tools. A connector is a plugin whose tools reach another service, for example a 3D model generator or your team's tracker. Plugins work with every AI choice: the Graspable agent, Claude Code and Codex get the same tools","s":"Connect your accounts of other services, and give the agent tools from other programs. How to add a plugin, how to write one, and the manifest format. A plugin gives the agent more tools. A connector is a plugin whose tools reach another service, for example a 3D model generator or your team's tracker. Plugins work with every AI choice: the Graspable agent, Claude Code and Codex get the same tools. Plugins work for projects on your computer. Cloud projects do not use them yet.","u":"/docs/plugins"},{"k":"api","t":"Plugins and connectors","h":"Add one","x":"Open Settings → Plugins and connectors . - Switch on a connector that comes with Graspable. Tick it, fill in what it asks for and choose Save . Then choose Test : it connects once and tells you how many tools it found. - Add an MCP server. Open Add a plugin or an MCP server , give it a name and its https address, and tick needs an access token if it does. Any service with an MCP server works this ","s":"Open Settings → Plugins and connectors . - Switch on a connector that comes with Graspable. Tick it, fill in what it asks for and choose Save . Then choose Test : it connects once and tells you how many tools it found. - Add an MCP server. Open Add a plugin or an MCP server , give it a name and its https address, and tick needs an access token if it does. Any service with an MCP server works this way. - Add a plugin. Paste the path of its folder on your computer, or the https address of its manifest. World Labs generates worlds with your own World Labs account. Meshy and Tripo generate and process models with your accounts at those services. UE-MCP for Unreal Engine gives the agent more tools in the Unreal Editor; it comes with the Pro plan and higher, works for Unreal projects only and has no Test button, because it needs a project. See Native XR. Secret values such as API keys are kept in your computer's keychain. They are handed to a run when it starts and are never written to a file by Graspable.","u":"/docs/plugins#add-one"},{"k":"api","t":"Plugins and connectors","h":"Characters and models","x":"In Settings → Plugins and connectors , enable Meshy or Tripo , enter your API key, then Save and Test . Testing lists the available tools; it does not generate a model or validate your balance. Generation and rigging use the provider's balance and follow your Graspable approval settings. With Meshy connected, ask the agent to find your completed API models, generate a character, rig it, or apply a","s":"In Settings → Plugins and connectors , enable Meshy or Tripo , enter your API key, then Save and Test . Testing lists the available tools; it does not generate a model or validate your balance. Generation and rigging use the provider's balance and follow your Graspable approval settings. With Meshy connected, ask the agent to find your completed API models, generate a character, rig it, or apply animation clips. Meshy's automatic rigging is intended for textured humanoids with clearly defined limbs. A generated model still needs a visual check for deformation, scale and foot contact. See Meshy's rigging requirements. With Tripo connected, ask the agent to create a character with a rigging step, or retrieve a completed API task by its ID. Its connector can inspect the CLI's recent task history; that is not a browser of your complete Tripo website library. Tripo offers multiple rig types, but suitability depends on the model and processing options. See Tripo rigging. Bring a model you already have 1. Export the model from its service, including its skeleton and animation clips. GLB is the simplest choice for web and XR projects because it can keep geometry, textures, skins and animations in one file. FBX is also useful for native engine projects. 2. In your desktop project, open Assets → Add and choose the exported file. Graspable keeps the original and copies it into your project. An existing file with the same name is preserved; a number is added to the new filename. 3. Select a GLB or glTF model in Assets to see its skin and joint counts, morph targets and named animation clips. The small preview shows the model at rest. Ask the agent to place the model in the scene and wire the clips to the behavior or sequence you want. For a .gltf export, keep its buffer and texture files beside it in the folders supplied by the exporter. Add the .gltf file; Graspable copies those local dependencies together into a model folder. References outside that folder or to remote addresses are refused. Export a GLB if the exporter uses a more complicated layout. Imports are limited to 300 MB, including glTF dependencies. Meshy API assets and the website's My Assets are separate. To bring a website model into Graspable, export it from Meshy and add the file; connecting an API key does not synchronize that library. See Meshy's account and download notes. Existing exports from Luma Genie and other services can be added the same way. There is no Genie generation or account-library connector here: Luma's currently documented public API provides image and video generation. A static model export does not gain a rig just by importing it. USD, USDA, USDC and USDZ assets open the OpenUSD inspector, which can build a GLB preview of supported scene data while preserving the original files. The connectors give the agent provider tools. There is no automatic library synchronization, universal retargeter, or skeletal clip editor in the animation timeline yet. Downloaded provider assets remain editable project files; keep a copy before remeshing a rigged character, because remeshing can discard its rig.","u":"/docs/plugins#characters-and-models"},{"k":"api","t":"Plugins and connectors","h":"What the agent may do with them","x":"Every plugin tool counts as reaching outside your project. With the default approval setting the agent asks you before it uses one. A plugin can mark a tool as read-only; those run without asking. See Approvals and safety. The tools appear in a run's Context list under Plugins , together with any plugin that could not connect and why.","u":"/docs/plugins#what-the-agent-may-do-with-them"},{"k":"api","t":"Plugins and connectors","h":"Connected services","x":"Open Settings → Connected services to connect your accounts of other services: Notion, Slack, GitHub, Google Sheets, Linear and thousands more. Find the service, choose Connect , and sign in at the service itself in your browser. When you come back, the service is in the list. The same list is on the website: sign in and open Account → Connectors to connect a service or disconnect one. Then add it","s":"Open Settings → Connected services to connect your accounts of other services: Notion, Slack, GitHub, Google Sheets, Linear and thousands more. Find the service, choose Connect , and sign in at the service itself in your browser. When you come back, the service is in the list. The same list is on the website: sign in and open Account → Connectors to connect a service or disconnect one. Then add it to the projects that need it. In a project, next to the box where you write to the agent, choose Connect apps and pick from the services you connected; Connect another app in that list takes you to Settings. From your next request on, the agent has that service's tools in this project: it can read a Notion page, post to Slack or open a GitHub issue while it builds. Remove a service from the project with the × beside its name; it stays connected for your other projects. - You sign in at the service. Graspable never sees your password, and the sign-in is kept by Pipedream, the company that runs the connections for Graspable. - A project gets only the services you added to it, so the agent is not handed tools the project has no use for. - The tools reach only the accounts you connected yourself. Disconnect (the bin icon) removes the sign-in for good. - Like every tool that reaches outside your project, the agent asks before it uses one. See Approvals and safety. - Connected services come with the Builder plan and higher; see Plans and billing. They work for projects on your computer. - A published app can use a connected service too, as you, through a grant you make for that app: see Connectors in your app.","u":"/docs/plugins#connected-services"},{"k":"api","t":"Plugins and connectors","h":"Write a plugin","x":"A plugin is a folder with a file named graspable-plugin.json : Key Meaning --- --- name Lowercase letters, digits and dashes. Tools are offered as <name <tool . kind plugin or connector . fields What the person fills in. secret: true goes to the keychain. default and optional are allowed. mcp Where the tools come from. ${NAME} is replaced with a field's value. risk What a tool can do: read , write","s":"A plugin is a folder with a file named graspable-plugin.json : Key Meaning --- --- name Lowercase letters, digits and dashes. Tools are offered as <name <tool . kind plugin or connector . fields What the person fills in. secret: true goes to the keychain. default and optional are allowed. mcp Where the tools come from. ${NAME} is replaced with a field's value. risk What a tool can do: read , write , exec or external . Tools not listed are external and ask first. skills A folder of skills the plugin brings. Default: skills . The mcp part has two forms. An address. type: \"http\" with an https url and optional headers . Two extras: - oauth : { \"tokenUrl\", \"clientId\", \"clientSecret\" } gets a bearer token with the client-credentials flow before connecting. - each : { \"name\": \"APP\", \"from\": \"APPS\" } opens one connection for every comma-separated item of the field APPS , with the item available as ${APP} . A program. type: \"stdio\" with a command , args and env . Graspable starts it in the plugin's folder and talks MCP over its standard input and output. node , npx and bun work even when Node.js is not installed, because Graspable's built-in runtime answers to those names. A plugin that starts a program must be added from a folder on your computer. A manifest added from an address may only point at an MCP address.","u":"/docs/plugins#write-a-plugin"},{"k":"api","t":"Plugins and connectors","h":"From a terminal","x":"For a run started with grasp run , secret fields come from the environment: GRASPABLE PLUGIN <NAME <FIELD , for example GRASPABLE PLUGIN MESHES API KEY . See CLI.","u":"/docs/plugins#from-a-terminal"},{"k":"api","t":"Skills","x":"Teach the agent how you do one kind of task, in your own words. Skills work with every AI choice. A skill is a folder with a file named SKILL.md : instructions for one kind of task. The agent sees the list of your skills and reads one when a task calls for it. Skills work the same with the Graspable agent, Claude Code and Codex. Use a skill for things you would otherwise repeat in every request: h","s":"Teach the agent how you do one kind of task, in your own words. Skills work with every AI choice. A skill is a folder with a file named SKILL.md : instructions for one kind of task. The agent sees the list of your skills and reads one when a task calls for it. Skills work the same with the Graspable agent, Claude Code and Codex. Use a skill for things you would otherwise repeat in every request: how your studio lights a scene, your naming rules, the steps to export for a client.","u":"/docs/skills"},{"k":"api","t":"Skills","h":"Find a skill","x":"Browse the library on the website. In the current desktop release, download a skill folder from its original source and add it in Settings. The in-app library, project-specific adding and removing, and recorded skill usage require Graspable 0.19.0 or later. Update your installed app to use them. In that update, open Skills from the Projects page, the project's Skills tab, or Settings → Plugins and","s":"Browse the library on the website. In the current desktop release, download a skill folder from its original source and add it in Settings. The in-app library, project-specific adding and removing, and recorded skill usage require Graspable 0.19.0 or later. Update your installed app to use them. In that update, open Skills from the Projects page, the project's Skills tab, or Settings → Plugins and skills. Search by a task, author or tool, and filter by XR & AR, 3D, animation, AI glasses, characters, Gaussian splats, games, film, performance or accessibility. Open a skill to read its original instructions, source, license, requirements and reviewed version. Choose Add skill in the app. Graspable checks the downloaded files against its reviewed catalog and adds any listed skill dependencies. It never overwrites a personal skill with the same name. Adding a skill does not execute scripts, install runtime packages or buy services. Existing skills stay usable offline; reading or adding a catalog entry needs a connection. Catalog versions come with app updates. The skills work with all supported agents on this computer. Switch them on or off for a project in its Skills tab. Native tools, hardware, accounts and runtime software still have their own requirements. The catalog is not a guarantee of device compatibility or generated-project quality. Entries marked Research reference depend on a larger upstream setup and cannot be added as incomplete packages.","u":"/docs/skills#find-a-skill"},{"k":"api","t":"Skills","h":"Projects using a skill","x":"A skill's page shows up to five public projects ordered by views or likes. Show more projects using this skill opens the public projects directory with that skill selected. You can change the skill filter there. Graspable records a catalog skill after an agent successfully reads its unchanged instructions through use skill . Having a skill installed or switched on does not count as using it. The n","s":"A skill's page shows up to five public projects ordered by views or likes. Show more projects using this skill opens the public projects directory with that skill selected. You can change the skill filter there. Graspable records a catalog skill after an agent successfully reads its unchanged instructions through use skill . Having a skill installed or switched on does not count as using it. The next completed publish includes those recorded skill IDs. Existing projects are not assigned skills retrospectively, and private, removed or unpublished projects are excluded. An empty list means no public projects have recorded that skill yet. This is contribution history, not a certification of the result.","u":"/docs/skills#projects-using-a-skill"},{"k":"api","t":"Skills","h":"Write one","x":"- name : lowercase letters, digits and dashes. - description : one or two sentences that say what the skill covers and when to use it. The agent decides from this line alone, so be specific. - The rest is the instructions. Other files in the folder can be referred to by name; the agent reads them when the instructions point to them.","u":"/docs/skills#write-one"},{"k":"api","t":"Skills","h":"Add it","x":"On a library page, follow Original skill to its source. Save the skill folder, including SKILL.md , its license and any companion files it references. For a standalone skill, save Read as Markdown as SKILL.md inside a folder. Read its requirements before adding it; some skills need other skills or software too. Open Settings → Plugins and connectors → Skills , paste the folder's path and choose Ad","s":"On a library page, follow Original skill to its source. Save the skill folder, including SKILL.md , its license and any companion files it references. For a standalone skill, save Read as Markdown as SKILL.md inside a folder. Read its requirements before adding it; some skills need other skills or software too. Open Settings → Plugins and connectors → Skills , paste the folder's path and choose Add skill . Graspable copies the folder into ~/.graspable/skills . From a terminal: A plugin can bring skills too. When your own skill has the same name as a plugin's, yours is used.","u":"/docs/skills#add-it"},{"k":"api","t":"Skills","h":"Skills that come with a template","x":"A template can come with skills of its own. The Unreal Engine template does: Graspable's skills for naming and folders, for keeping a scene fast on a headset and for making it look like professional work, and two public collections for Unreal's own systems (C++, Blueprints, materials, input, UI, packaging and more). Every web template comes with a security review skill (the public OWASP skill, MIT","s":"A template can come with skills of its own. The Unreal Engine template does: Graspable's skills for naming and folders, for keeping a scene fast on a headset and for making it look like professional work, and two public collections for Unreal's own systems (C++, Blueprints, materials, input, UI, packaging and more). Every web template comes with a security review skill (the public OWASP skill, MIT licence): ask the agent whether your app is safe to publish, and it reviews the project against the OWASP Top 10 and the ASVS, from what the app does with addresses and files to what it sends to connectors. They are used in every project made from that template, with nothing to set up. Settings → Plugins and connectors → Skills lists them under the template's name, each with a switch. Switch one off to leave it out of your projects. To replace one, add a skill of your own with the same name: yours is used instead.","u":"/docs/skills#skills-that-come-with-a-template"},{"k":"api","t":"Skills","h":"Skills in one project","x":"Open Skills beside the conversation, then Add skills . Search the directory without leaving your scene, choose a skill, check its requirements and choose Add to project . The full instructions and source are one click away. The full directory keeps your project selected and has a Back to project link. Skills added here, including their listed dependencies, belong only to this project. Remove leave","s":"Open Skills beside the conversation, then Add skills . Search the directory without leaving your scene, choose a skill, check its requirements and choose Add to project . The full instructions and source are one click away. The full directory keeps your project selected and has a Back to project link. Skills added here, including their listed dependencies, belong only to this project. Remove leaves a skill out of its next request without changing another project. To restore it, choose Add skills , then Add under Available on this computer . Removing a skill does not remove models or other work it helped create. Your shared skills, plugin skills and template skills also appear here. A template skill switched off in Settings stays unavailable until you switch it on there. Settings remains the place to install or remove shared skills for all projects.","u":"/docs/skills#skills-in-one-project"},{"k":"api","t":"Skills","h":"Work with local Blender","x":"If Blender is already installed, add Blender Core , then the skills for the work: materials, animation, lighting, exporting or asset checks. Ask the agent to use your local Blender, keep an editable .blend source and export a .glb for the experience. You can keep editing the source in Blender and ask the agent to regenerate the exported asset afterward. The agent uses Blender's local scripting int","s":"If Blender is already installed, add Blender Core , then the skills for the work: materials, animation, lighting, exporting or asset checks. Ask the agent to use your local Blender, keep an editable .blend source and export a .glb for the experience. You can keep editing the source in Blender and ask the agent to regenerate the exported asset afterward. The agent uses Blender's local scripting interface through its normal command approvals. A skill does not install Blender or require a paid rendering service. Procedural materials and constraints may need baking before export; check the exported scene in Graspable, as a Blender render alone does not prove how it looks or performs in the browser or a headset.","u":"/docs/skills#work-with-local-blender"},{"k":"api","t":"Skills","h":"How the agent uses them","x":"The names and descriptions of your skills are part of every run's prompt. When a task matches one, the agent reads the skill and follows it. You can see this in the run as a use skill step, and the list itself under Context . Skills that already follow this layout for other tools, a folder with a SKILL.md that has name and description , can be added as they are.","u":"/docs/skills#how-the-agent-uses-them"},{"k":"api","t":"CLI","x":"Use Graspable from a terminal or a script. Create projects, start runs, follow them, check and preview. grasp is Graspable as a command. It does what the app does: create projects, ask the agent to build, follow the run, check the result and start the preview.","u":"/docs/cli"},{"k":"api","t":"CLI","h":"Install it","x":"macOS and Linux Windows Homebrew (macOS) The installer downloads the build for your computer, checks it against its published checksum and puts the grasp command in ~/.local/bin (on Windows, in a folder it adds to your PATH ). Nothing else needs installing: grasp brings what it needs to install packages, build and run a preview. If your terminal says grasp: command not found , the installer's last","s":"macOS and Linux Windows Homebrew (macOS) The installer downloads the build for your computer, checks it against its published checksum and puts the grasp command in ~/.local/bin (on Windows, in a folder it adds to your PATH ). Nothing else needs installing: grasp brings what it needs to install packages, build and run a preview. If your terminal says grasp: command not found , the installer's last lines show the one line that adds ~/.local/bin to your PATH . Run it and open a new terminal. Do you already have the Graspable app? Then you can also install the command from Settings → Plugins and connectors → Use Graspable from other tools . Both ways give the same grasp .","u":"/docs/cli#install-it"},{"k":"api","t":"CLI","h":"Choose how to edit","x":"If a connected coding agent will edit the files itself, it can use Graspable through MCP to create a free web project, read and write files, build, check and preview it. Those operations need no additional AI sign-in and spend no Graspable Credits. The connected agent’s own usage terms still apply. Premium templates and services keep their plan requirements. The installer links to the connection g","s":"If a connected coding agent will edit the files itself, it can use Graspable through MCP to create a free web project, read and write files, build, check and preview it. Those operations need no additional AI sign-in and spend no Graspable Credits. The connected agent’s own usage terms still apply. Premium templates and services keep their plan requirements. The installer links to the connection guide. To ask Graspable’s agent to generate changes with grasp run , configure its AI account first.","u":"/docs/cli#choose-how-to-edit"},{"k":"api","t":"CLI","h":"Sign in","x":"Sign in once: It asks what should power the agent and signs you in: Choice What it uses Also directly --- --- --- ChatGPT subscription The Graspable agent, signed in with ChatGPT in your browser grasp login chatgpt Claude subscription Claude Code on this computer, with its own sign-in grasp login claude ChatGPT through Codex The Codex CLI on this computer, with its own sign-in grasp login codex An","s":"Sign in once: It asks what should power the agent and signs you in: Choice What it uses Also directly --- --- --- ChatGPT subscription The Graspable agent, signed in with ChatGPT in your browser grasp login chatgpt Claude subscription Claude Code on this computer, with its own sign-in grasp login claude ChatGPT through Codex The Codex CLI on this computer, with its own sign-in grasp login codex An API key The Graspable agent with a key from Anthropic, OpenAI, OpenRouter and others grasp login key anthropic An API key is typed once and kept in your computer's keychain (on Linux, in the Secret Service). On a computer without one, set the key in the environment variable GRASPABLE API KEY instead. Check what is set up at any time: Change the default later with grasp use , for example grasp use codex or grasp use pi --provider openrouter --model anthropic/claude-sonnet-4.5 . Shared with the Graspable app grasp and the app use the same set-up on one computer: - Projects, runs, memory, plugins and skills are the same in both. - The default AI and model are shared. Change it in the app's Settings or with grasp use , and the other follows. - Sign-ins with a subscription are shared: a ChatGPT sign-in made in the app works in grasp and the other way round, and Claude Code and Codex keep their own sign-in, which both use. - API keys are not shared. The app keeps its keys in its own keychain entry, which grasp cannot read. If the app's default is an API key account, add that key once with grasp login key <provider . When the app is open, grasp talks to it, so a run you start in the terminal appears in the app as it happens. When it is closed, grasp starts the same agent in the background. On a computer without a screen, you can also start it yourself with grasp serve .","u":"/docs/cli#sign-in"},{"k":"api","t":"CLI","h":"Your first run","x":"1. grasp new creates a project and prints its id. 2. grasp run sends your request. It prints what the agent does and returns when the run is finished. The exit code is 0 when the run completed. Read the check report too: a browser or native runtime check can be skipped when its environment is unavailable. A completed run alone does not prove the scene ran. 3. grasp preview starts the live preview.","s":"1. grasp new creates a project and prints its id. 2. grasp run sends your request. It prints what the agent does and returns when the run is finished. The exit code is 0 when the run completed. Read the check report too: a browser or native runtime check can be skipped when its environment is unavailable. A completed run alone does not prove the scene ran. 3. grasp preview starts the live preview. Open the address in a browser. When the agent wants to run a command or reach outside the project, grasp run asks you first. See Approvals below.","u":"/docs/cli#your-first-run"},{"k":"api","t":"CLI","h":"Update","x":"Run the install line again. It installs the newest version and removes the old one. With Homebrew: grasp version prints the installed version.","u":"/docs/cli#update"},{"k":"api","t":"CLI","h":"Uninstall","x":"Delete ~/.local/bin/grasp and the folder ~/.graspable/cli . With Homebrew: brew uninstall --cask grasp . Your projects and settings stay where they are.","u":"/docs/cli#uninstall"},{"k":"api","t":"CLI","h":"Commands","x":"Command What it does --- --- grasp login , grasp whoami , grasp use , grasp logout Sign in and choose the default AI grasp projects List projects grasp new <name [--template <id ] Create a project. grasp templates lists the templates. A premium template needs your Graspable session in GRASPABLE SESSION , so your plan can be checked grasp files <project , grasp cat <project <path Look at a project'","s":"Command What it does --- --- grasp login , grasp whoami , grasp use , grasp logout Sign in and choose the default AI grasp projects List projects grasp new <name [--template <id ] Create a project. grasp templates lists the templates. A premium template needs your Graspable session in GRASPABLE SESSION , so your plan can be checked grasp files <project , grasp cat <project <path Look at a project's files grasp run <project \"<request \" Ask the agent to build or change something grasp runs <project Earlier runs grasp logs <run [--follow] What a run did grasp approve <run <id yes\\ no Answer an approval request grasp cancel <run Stop a run grasp check <project Build the project and load it in a headless browser grasp preview <project Start the live preview and print its address grasp history <project The project's versions, newest first: one per run that changed files grasp restore <project <version Put the files back to a version. Recorded as a new version, so nothing is lost grasp publish <project [--title \"…\"] [--name <name ] [--private] [--no-fork] Publish the app at an address of its own. Needs your Graspable session in GRASPABLE SESSION (the app signs in for you) grasp fork <app-id Continue from a public app: a new project with its source and history grasp plugins , grasp skills Manage plugins and skills grasp mcp The Graspable MCP server grasp token create <name [--scope read\\ run\\ admin] An access token for the API grasp serve [--port <n ] Run the agent without the app grasp status Is the agent running? Add --json to projects , runs , logs , history , restore and run for output a program can read. With --json , grasp run prints one event per line.","u":"/docs/cli#commands"},{"k":"api","t":"CLI","h":"Options of grasp run","x":"Option Meaning --- --- --engine claude-code\\ codex\\ pi What powers this run. Default: what you set up with grasp login --model <id The model. Optional for claude-code and codex --provider <id For pi : the provider, for example anthropic , openai or openrouter --policy auto\\ guarded\\ strict What the agent asks first. Default: guarded --approve yes\\ no Answer every approval request the same way --no","s":"Option Meaning --- --- --engine claude-code\\ codex\\ pi What powers this run. Default: what you set up with grasp login --model <id The model. Optional for claude-code and codex --provider <id For pi : the provider, for example anthropic , openai or openrouter --policy auto\\ guarded\\ strict What the agent asks first. Default: guarded --approve yes\\ no Answer every approval request the same way --no-wait Print the run id and return at once claude-code and codex use the sign-in of those tools on your computer. pi is the Graspable agent, with your ChatGPT sign-in or an API key.","u":"/docs/cli#options-of-grasp-run"},{"k":"api","t":"CLI","h":"Approvals","x":"With the guarded policy the agent asks before it runs a command or reaches outside the project. In a terminal, grasp run asks you. In a script it waits; answer from another process with grasp approve , or pass --approve yes , or use --policy auto to let the agent work without asking. See Approvals and safety.","u":"/docs/cli#approvals"},{"k":"api","t":"CLI","h":"Usage counts","x":"grasp counts which commands and MCP tools are used, by which agent, and whether they worked: never their arguments or output. See Privacy and telemetry for exactly what is sent. Turn it off with grasp telemetry off (or DO NOT TRACK=1 ); grasp telemetry says whether it is on.","u":"/docs/cli#usage-counts"},{"k":"api","t":"CLI","h":"Long-running and background work","x":"grasp logs --follow returns when the run is finished, with the same exit code as grasp run .","u":"/docs/cli#long-running-and-background-work"},{"k":"api","t":"MCP","x":"Let another AI agent build with Graspable. Set-up for Claude Code, Codex and other MCP clients, and the list of tools. Graspable has an MCP server. With it, another AI agent can create projects, ask Graspable's agent to build, follow the run, read and write files, check the result and start the preview.","u":"/docs/mcp-server"},{"k":"api","t":"MCP","h":"Set it up","x":"Install grasp through Settings → Plugins and connectors → Use Graspable from other tools in the desktop app. Agent-operated installation is covered in CLI. Direct file editing and local checks work without signing in to another AI account. Configure Graspable's AI account only if you want to use start run to delegate generation. Claude Code Codex Any other MCP client that starts a program: Over HT","s":"Install grasp through Settings → Plugins and connectors → Use Graspable from other tools in the desktop app. Agent-operated installation is covered in CLI. Direct file editing and local checks work without signing in to another AI account. Configure Graspable's AI account only if you want to use start run to delegate generation. Claude Code Codex Any other MCP client that starts a program: Over HTTP. The same server is at POST /mcp of the local API, with an access token: Browser checks For web projects, Graspable checks with Chrome for Testing, downloaded from Google into its own runtime folder. On Windows and Linux it uses Chrome, Edge, Chromium or Brave already on your computer and downloads Chrome for Testing only when none is there; on a Mac it downloads it in any case and uses your own Chrome until the download is done, because an installed Chrome tries to update its own app when a check starts it. The first check waits for that download. It does not use your browser profile. If the download or browser launch fails, the report can contain a skipped check even when the build passes. Read the browser result before treating the scene as tested. In the desktop app, Settings → Components shows the browser and its download status. See check screenshots for returning the available picture to your agent. Next desktop release: browser checks on macOS use a separate test keychain, avoiding login-keychain prompts during headless startup. This fix is not included in 0.19.0.","u":"/docs/mcp-server#set-it-up"},{"k":"api","t":"MCP","h":"Tools","x":"Tool What it does --- --- inspect motion Validate a motion JSON file and inspect its pose at a frame; read-only, experimental. See motion inspection list templates The project templates list projects The person's projects create project Create a free web project. Create premium or native projects in the app first list files , read file , write file Work with a project's files start run Ask Graspab","s":"Tool What it does --- --- inspect motion Validate a motion JSON file and inspect its pose at a frame; read-only, experimental. See motion inspection list templates The project templates list projects The person's projects create project Create a free web project. Create premium or native projects in the app first list files , read file , write file Work with a project's files start run Ask Graspable's agent to build or change something, in plain words get run What a run has done so far. Waits for it to finish or to need an answer approve Answer a run's approval request cancel run Stop a run list runs Earlier runs of a project check project Build and check a project with its browser or native engine project history The project's versions, newest first: one per run that changed files restore project Put the project's files back to a version. Recorded as a new version start preview Start a web project’s live preview and return its local address Version 0.17.2 also includes read usd and write usd for scoped USD inspection, separate edit layers and supported scene conversion. read motion , edit motion , export motion and read otio edit cover animation takes and editorial handoffs. inspect film lists delivery summaries or reads exact video revisions; encode spatial video creates a native Apple master and returns a small receipt without movie bytes. See film delivery. Use Graspable 0.17.2 or later for these tools. See OpenUSD scenes and the animation API.","u":"/docs/mcp-server#tools"},{"k":"api","t":"MCP","h":"Edit with your connected agent","x":"1. Use list projects to find the project, or create the free react-xr-developer starter with create project . 2. Wait for list projects to report install: done . An installation failure needs attention before editing or building. 3. Use list files and read file to inspect the source. write file takes the complete replacement file, with a project-relative path. Preserve unrelated edits and coordina","s":"1. Use list projects to find the project, or create the free react-xr-developer starter with create project . 2. Wait for list projects to report install: done . An installation failure needs attention before editing or building. 3. Use list files and read file to inspect the source. write file takes the complete replacement file, with a project-relative path. Preserve unrelated edits and coordinate with anyone else editing that file. 4. Call check project . It builds a web project and loads it in a headless browser. Correct any errors and check again. 5. Call start preview , then inspect the running scene. Return the preview address and explain any device checks still needed. The connected agent supplies the edits through its existing AI account. This flow does not call Graspable's model or spend Graspable Credits. Template entitlements and any separately requested paid services still apply. Testing the intended interaction is a separate step from a successful build and browser load. Try a small scene edit With the free React XR starter open, ask your connected agent: Change the New ball button to New pair. Each press should add two balls, keep the existing limit of eight, and show the current ball count on the sign. Tidy up should still leave one ball. Edit the files yourself, then use Graspable to check and preview the result. After the check passes, try the interaction: the count starts at one, becomes three after one press, stops at eight after repeated presses, and returns to one with Tidy up . This exercises source editing, scene state and a control that already belongs to the template. It needs no new assets or generation service. This example was checked with Graspable 0.18.1 on macOS, including those mouse interactions in a desktop browser. It is not a headset or phone AR test.","u":"/docs/mcp-server#edit-with-your-connected-agent"},{"k":"api","t":"MCP","h":"Work on an existing native project","x":"Create a Unity, Unreal Engine or Apple Vision Pro project in the desktop app with a plan that includes native XR. This supplies the plan authorization that create project does not currently obtain through MCP. Keep the corresponding engine tools installed on the same computer. Your connected agent can then select the project with list projects , read and edit its source, and call check project . T","s":"Create a Unity, Unreal Engine or Apple Vision Pro project in the desktop app with a plan that includes native XR. This supplies the plan authorization that create project does not currently obtain through MCP. Keep the corresponding engine tools installed on the same computer. Your connected agent can then select the project with list projects , read and edit its source, and call check project . The check uses the native engine: Unity play mode, the visionOS Simulator or Unreal's desktop game view. Read the summary; a skipped run is not a successful runtime test. Open the project's preview in Graspable to inspect the picture. start preview is for web projects and cannot return a browser address for a native app. These checks do not verify hand input, room tracking, comfort or performance on a physical headset. See Native XR for engine setup and device testing. Check screenshots (0.19.0 and later) Graspable 0.19.0 and later supports an optional includeImage: true argument to check project . It returns the available screenshot as an MCP image alongside the text report, so your connected agent can inspect the result. Graspable 0.18.1 does not return that image yet. The screenshot is sent to the connected agent only when requested. A runtime failure can still include a useful picture. Missing or oversized images produce a note; a text-only result is not visual verification. An unchanged project may reuse its previous check and picture, which the summary identifies. Test the intended interaction separately: the image does not prove that a button or gesture works. The same release adds engine kind and required plan feature to list templates , engine kind to list projects , and guidance for direct editing when an agent connects. No additional model sign-in is needed for direct edits or checks.","u":"/docs/mcp-server#work-on-an-existing-native-project"},{"k":"api","t":"MCP","h":"Delegate to Graspable's agent","x":"Configure its AI account first in Settings → AI for the agent or through grasp login . Delegated runs use that account's allowance or balance. 1. create project , or list projects to find one. 2. start run with the request. It returns a run id at once. 3. get run with that id until the state is completed , failed or cancelled . Pass the since number from the last answer to get only what is new. 4.","s":"Configure its AI account first in Settings → AI for the agent or through grasp login . Delegated runs use that account's allowance or balance. 1. create project , or list projects to find one. 2. start run with the request. It returns a run id at once. 3. get run with that id until the state is completed , failed or cancelled . Pass the since number from the last answer to get only what is new. 4. When get run lists needs approval , ask the person. Call approve only when they have authorized that action; otherwise leave the request for them to answer in Graspable. 5. For a web project, use start preview to look at the result. For a native project, open its preview in Graspable. A web run uses build and browser checks; a native run uses its engine checks. Read the reported outcome, including any skipped checks. A failed check goes back to Graspable's agent, up to three times, before the run fails.","u":"/docs/mcp-server#delegate-to-graspable-39-s-agent"},{"k":"api","t":"MCP","h":"What the other agent may do","x":"Each tool call is counted by the tool's name and the client's name, never with its arguments. See Privacy and telemetry; grasp telemetry off stops it. grasp mcp acts with the run level of access: it can create projects, edit files, start runs and answer approvals. It cannot change plugins, sign-ins or settings, and it cannot delete projects. Approval requests exist so that a person decides. An age","s":"Each tool call is counted by the tool's name and the client's name, never with its arguments. See Privacy and telemetry; grasp telemetry off stops it. grasp mcp acts with the run level of access: it can create projects, edit files, start runs and answer approvals. It cannot change plugins, sign-ins or settings, and it cannot delete projects. Approval requests exist so that a person decides. An agent that calls approve decides in your place. If you do not want that, tell the agent to stop and ask you when a run needs an approval, or start runs with the strict policy and answer in the app.","u":"/docs/mcp-server#what-the-other-agent-may-do"},{"k":"api","t":"API","x":"The HTTP API of the Graspable agent on your computer. Access tokens and their levels, and the requests for projects, files and runs. The Graspable app talks to its agent over an HTTP API on your computer. Scripts and other programs can use the same API. It listens on 127.0.0.1 only. For most uses the grasp command or the MCP server is simpler. Use the API when you want to build your own tool.","u":"/docs/api"},{"k":"api","t":"API","h":"Address and access","x":"The port changes each time the agent starts, so ask for the address each time, or read it from ~/.graspable/agent/api.json . Every request needs Authorization: Bearer <token . Create a token in Settings → Plugins and connectors → Access tokens , or: The token is shown once. Graspable stores only a hash of it. Level May --- --- read Look at projects, files, runs and usage run Also create projects, ","s":"The port changes each time the agent starts, so ask for the address each time, or read it from ~/.graspable/agent/api.json . Every request needs Authorization: Bearer <token . Create a token in Settings → Plugins and connectors → Access tokens , or: The token is shown once. Graspable stores only a hash of it. Level May --- --- read Look at projects, files, runs and usage run Also create projects, edit files, start and steer runs, answer approvals, start previews, use memory admin Also change plugins, skills, components, sign-ins and tokens, move projects to the cloud, delete projects A request above the token's level is answered with 403 .","u":"/docs/api#address-and-access"},{"k":"api","t":"API","h":"Projects and files","x":"Request Does --- --- GET /projects List projects POST /projects { \"name\", \"template\" } Create a project. Packages install in the background; install becomes done GET /projects/:id One project GET /projects/:id/files File paths GET /projects/:id/file?path=App.jsx { \"path\", \"content\" } PUT /projects/:id/file { \"path\", \"content\" } Write a file POST /projects/:id/preview Start the preview: { \"url\", \"p","s":"Request Does --- --- GET /projects List projects POST /projects { \"name\", \"template\" } Create a project. Packages install in the background; install becomes done GET /projects/:id One project GET /projects/:id/files File paths GET /projects/:id/file?path=App.jsx { \"path\", \"content\" } PUT /projects/:id/file { \"path\", \"content\" } Write a file POST /projects/:id/preview Start the preview: { \"url\", \"port\" } DELETE /projects/:id/preview Stop the preview POST /projects/:id/check Build and check with the web browser or native engine: { \"pass\", \"build\", \"browser\" } . The browser field holds the runtime result for either kind POST /projects/:id/install Install the project's packages again POST /projects/:id/publish { \"accessToken\", \"title\", \"visibility\", \"forkable\" } Publish the app; accessToken is a Graspable session POST /projects/fork { \"appId\" } A new project from a public app's source GET /projects/:id/history , GET /projects/:id/history/:sha , POST /projects/:id/history/:sha/restore The project's versions, one version's changes, and going back to one Paths are relative to the project. Paths that leave the project are refused. Check images (0.19.0 and later) A passing build does not mean the app ran. Read the runtime summary in browser for skipped checks, even when pass is true. Version 0.19.1 and later also returns browser.skipped with the reason when a browser, visionOS simulator, Unity play-mode connection or Unreal Metal toolchain is unavailable. Skipped native checks are retried instead of cached. Version 0.19.0 reports the reason in its summary only. If your integration requires a rendered result, reject a skipped check and inspect the available image separately. Graspable 0.19.0 and later accepts POST /projects/:id/check?includeImage=true to include the check's available screenshot as image: { mimeType: \"image/png\", data: \"<base64 \" } plus an imageNote . Graspable 0.18.1 returns text results only. Omit the parameter for a compact response. The PNG is limited to 4 MiB; missing, invalid or larger pictures are omitted with a note. A failing runtime check may still carry a picture. Cached checks may reuse their picture; inspect the summary. This uses the existing run-level check permission and does not grant access to settings or credentials.","u":"/docs/api#projects-and-files"},{"k":"api","t":"API","h":"Motion inspection (experimental)","x":"GET /projects/:id/motion?path=film.motion.json&frame=15 validates a motion document and returns its pose at that frame. It needs read access. It does not play, render or change the scene. Preview editing is separate from this read-only endpoint; see animation editing. A document contains version: 1 , fps: { numerator, denominator } , durationFrames and tracks . Each track has a unique id , a stabl","s":"GET /projects/:id/motion?path=film.motion.json&frame=15 validates a motion document and returns its pose at that frame. It needs read access. It does not play, render or change the scene. Preview editing is separate from this read-only endpoint; see animation editing. A document contains version: 1 , fps: { numerator, denominator } , durationFrames and tracks . Each track has a unique id , a stable object target , a property and keys . Properties are position , scale , quaternion , opacity and visible . A target can have only one track for each property. Each key contains frame , value and interpolation ( linear , hold , smooth or bezier , applied until the next key). Smooth applies cubic smoothstep easing. Bézier keys also require bezier: [x1, y1, x2, y2] : both x values must be between 0 and 1, and y values between −4 and 4. Opacity and morph channels restrict y to 0–1 to avoid out-of-range weights. Other interpolation modes must omit bezier . The sampler solves the curve’s time coordinate before evaluating its value. Reads, copies and edits preserve these controls; USD export bakes nonlinear timing at whole frames. Frames are unique, increasing integers from zero through durationFrames - 1 . Inspection also accepts fractional frames within that range. Position and scale values have three components; quaternions have four in x, y, z, w order and are normalized. Visibility uses booleans with hold interpolation; opacity is between zero and one. Rotations follow the shortest arc. Before the first key and after the last, the nearest pose holds. Camera shots with an end pose accept optional interpolation and bezier controls. Omitted interpolation keeps smooth timing. The selected curve applies to position, rotation and field of view from startFrame to endFrame - 1 . Camera Bézier values must all stay between 0 and 1; other modes omit bezier . Shot edits preserve timing, and USD export bakes the resulting movement. At frame 15 this returns the character's position as [1, 0, 0] . Inspecting another frame first does not change that result. Invalid documents and frames return 400 with an explanation. The API accepts at most two million text characters, 4096 tracks and 100000 keys in total.","u":"/docs/api#motion-inspection-experimental"},{"k":"api","t":"API","h":"Animation editing API (in development)","x":"These endpoints are available from 0.17.2. The animation editor uses them to save a take without overwriting a newer version. Request Does --- --- GET /projects/:id/authoring/motion?path=film.motion.json Read { revision, document } ; both are null when the take does not exist PUT /projects/:id/authoring/motion { \"path\", \"revision\", \"document\" } Save a validated take using the revision returned by ","s":"These endpoints are available from 0.17.2. The animation editor uses them to save a take without overwriting a newer version. Request Does --- --- GET /projects/:id/authoring/motion?path=film.motion.json Read { revision, document } ; both are null when the take does not exist PUT /projects/:id/authoring/motion { \"path\", \"revision\", \"document\" } Save a validated take using the revision returned by the last read POST /projects/:id/authoring/operations Apply a revision-checked operation batch or save a draft, with a retry receipt and guarded undo GET /projects/:id/authoring/history?path=… Read recent accepted edit summaries, optionally filtered to a take GET /projects/:id/authoring/query?path=… Read a compact overview, selected data or a bounded change catalog POST /projects/:id/authoring/export Export a revision-checked take as .usda or .otio GET /projects/:id/authoring/otio Plan supported edits from an OTIO file without changing the take POST /projects/:id/authoring/bind { \"source\", \"signature\" } Give a rendered scene object a persistent animation identity, after checking its source has not changed Saving a full take through PUT and binding objects need admin access. The operations endpoint needs run access, which already permits editing project files. A successful write returns the saved data and a fragment update event. An outdated revision or an active agent run returns 409 ; reload and review before retrying. Edits must use a project path ending in .motion.json ; protected files and paths through symlinks cannot be saved through this endpoint. These checks do not provide remote collaborative sessions or participant roles. The operations endpoint accepts version: 1 , path , actor , a unique requestId , and the revision from the last read. Supply exactly one of: - edits : between 1 and 128 operations ( key , remove-key , move-key , remove-track , duration , shot , remove-shot , audio , or remove-audio ). - document : the complete validated draft, for publishing local edits as one change. - bulk : a copy recipe with sourceTarget , distinct targets , collision ( reject or replace ), and optional properties , staggerFrames , frameOffset , positionOffset , and random: { seed, frameJitter?, positionJitter? } . Random offsets are stable per target and seed; staggering follows destination order. The resulting tracks remain individually editable. The operation rejects out-of-range keys and the document's track/key limits before expanding. - undoOf : a previous request ID submitted by this actor with the same access credential. Use the same request ID and body when retrying a request whose response was lost. A replay returns the current snapshot, the original appliedRevision , replayed: true , and no repeated file write. Reusing an ID with different content returns 409 . Receipts survive an agent restart; recovery refuses to overwrite a take changed after an interrupted save. Up to 32 recent receipts are retained, bounded to 32 MB. Never reuse request IDs after that window. Undo checks the tracks, shots, sounds, frame rate and duration affected by the original change. It preserves unrelated edits and refuses to restore an item another edit has since changed. Actor names are labels within an access credential, not authenticated people. This supports cooperating clients on one project host; it is not a remote invitation or presence system. An active agent run still locks authoring writes. Outside agents can use the read motion , motion history and edit motion MCP tools for this workflow. A stale revision must be read and reviewed before a new batch is submitted. History returns accepted change summaries, caller labels, request IDs, revisions and timestamps (null for older receipts), without credentials or private snapshots. New built-in agent requests also receive bounded recent history automatically; current files remain authoritative. Read only what an edit needs read motion now defaults to a small overview: revision, frame rate, duration and counts. Its view can be tracks for channel identities and key counts, keys for one required track , shots , audio , or changes . Track catalogs support exact target , track and property filters; shots and audio support an exact id . Key reads accept inclusive fromFrame and toFrame and include neighboring keys needed for interpolation. A range may span several pages; collect them before evaluating the curve. Pages default to 32 items and allow limit from 1 to 256. Repeat the same filters with nextCursor ; a cursor is tied to those filters and the exact content revision. revision pins a read and rejects a newer file. ifRevision returns only an unchanged receipt if the file still matches. Use that only when the previous data remains available in your context. For view: \"changes\" , supply since with a previously read revision. The result lists added, updated and removed track, shot and audio identities, plus current counts and whether timing metadata changed. Read the affected items separately. If the older revision has left the bounded cache, resetRequired asks for fresh reads. Each request still checks the file on disk, including edits made outside Graspable. The edit motion tool returns a short receipt without echoing the film or file events. revision is the current file; appliedRevision identifies the original accepted edit, which can differ on a retry after another edit. Full desktop snapshots and file-update events remain available through the existing authoring API. Interchange files export motion accepts path , current revision , format ( usda or otio ) and output . USD supports an optional aspectRatio (default 16:9). OTIO supports name and a project-relative videoSource pointing to a full-film render. Save OTIO at the project root to keep media paths correct. The export endpoint requires run access. The tool returns the output path, content hash ( outputRevision ), byte count, source revision, cache status and capability warnings. File contents stay on disk. To replace a different existing output, provide its current outputRevision ; identical output is reused. Cached generation is keyed by source content, export options and adapter version. It does not skip revision or destination checks. USD exports animated target transforms, visibility and perspective cameras. It uses meters, Y up and exact rational frame timing, baking nonlinear curves at whole frames. Target transforms are local channels on placeholder objects: geometry, parent hierarchy, materials, rigs, audio and interactive code are not reconstructed. Opacity is a custom attribute. Fractional-frame behavior can differ from the original easing. This is an animation handoff, not a complete scene export or arbitrary USD import. OTIO exports camera-shot cuts, gaps and overlapping audio references. A render is needed to view shot media in another editor. Mix gain and fades are retained as metadata that other editors may not apply. plan otio import accepts path , current revision , an edited input file and optionally its expected inputRevision . It returns a reviewable batch for existing identities. Apply its exact edits with edit motion and the returned source revision. Empty edits mean nothing changed. The plan is a snapshot of that OTIO revision; later file changes require a new plan. Supported imports include shot renaming and retiming, audio timing and source offsets, and mute. Removing clips requires allowRemovals: true . Unsupported effects, transitions, new identities, media relinking or changed frame rates are rejected. Object keys stay at their original absolute frames, so inspect the animation after changing cut timing. Plans must fit one atomic batch of 128 operations; larger edits are refused. The unreleased headset editing connection uses three admin-only endpoints on the desktop host: Request Does --- --- POST /projects/:id/authoring/headset { \"path\": \"film.motion.json\" } Start a connection for this project and take; return its session ID, expiry and headset URL PUT /projects/:id/authoring/headset/:session Publish the desktop's sampled state and receive at most one validated headset intention DELETE /projects/:id/authoring/headset/:session Revoke the connection The headset receives a separate credential limited to this connection, never the desktop API credential. It expires after 30 minutes and is revoked when replaced or stopped. The preview exchanges URL credentials for HttpOnly cookies and removes them from the address. Headset writes require the preview's origin. If the desktop stops updating for ten seconds, headset editing stops accepting intentions until it reconnects. Mutations carry the state sequence they were made against; stale poses and saves are discarded instead of being applied at a different frame or to a different selection. Transport controls remain usable while playback advances. The desktop remains responsible for applying and saving every edit.","u":"/docs/api#animation-editing-api-in-development"},{"k":"api","t":"API","h":"USD scenes (in development)","x":"The USD inspector and these endpoints are available from 0.17.2. See OpenUSD scenes for the supported editing and conversion workflow. Request Does --- --- GET /usd/status Report whether this build has its managed OpenUSD runtime GET /projects/:id/usd?path=…&query=… Read a stage or a scoped page; query is a URL-encoded JSON object POST /projects/:id/usd Save a separate edit layer or converted asse","s":"The USD inspector and these endpoints are available from 0.17.2. See OpenUSD scenes for the supported editing and conversion workflow. Request Does --- --- GET /usd/status Report whether this build has its managed OpenUSD runtime GET /projects/:id/usd?path=…&query=… Read a stage or a scoped page; query is a URL-encoded JSON object POST /projects/:id/usd Save a separate edit layer or converted asset with a source revision check Reads require read access; writes require run access. The MCP tools are read usd and write usd , with the same request fields plus project . Pro/Studio can use POST /asset-library with { \"entitlement\": \"…\", \"query\": \"pallet\", \"offset\": 0 } to search NVIDIA SimReady and Scenario assets. Optional filters accepts provider ( nvidia or scenario ) and format ( usd , glb , blend , spz , ply , vox or fbx ). Supply id instead of query for source, license, supplied preview, format variants and import limits. Search returns 24 results per page, a nullable nextOffset and pinned source revisions . For a Scenario GLB, POST /asset-library with id , variant , preview: true and entitlement returns verified binary model data, bounded to 32 MiB. This is for the graphical viewer, not agent context. It does not write into the project. The signed grant is checked before downloading and before returning a cached or downloaded preview. A shared preview cache holds at most four files and 64 MiB; cancelled or failed downloads are not cached. POST /projects/:id/asset-library with { \"entitlement\": \"…\", \"id\": \"pallet a01\" } imports a checked NVIDIA package. For Scenario, use its returned id and a variant path from the detail response to download only that format; omitting variant selects the first supplied output. The receipt contains the imported path, attribution path, file count, size and revision. Both endpoints require run access and a current signed plan statement from /api/billing/entitlement ; the grant is checked again before files appear. Imports preserve source files and attribution, verify hashes and refuse files over 64 MiB. No model bytes are returned in agent tool responses. A read without query returns a stage overview, revision , rootRevision and a dependency count. revision covers the complete local dependency graph; rootRevision hashes only the root file. A glTF read provides a revision for conversion without pretending that it is already a USD stage. Use query.view to request stage , prims , prim , properties , samples , layers , variants or changes . Scope reads with prim , property and time . limit is 1–128, default 32. Continue with nextCursor and identical filters; a cursor is pinned to its revision. Supplying revision rejects a changed scene. Supplying ifRevision returns unchanged: true when the graph still matches. A stage read retains a bounded change baseline. For view: \"changes\" , set since to that revision to receive added, changed and removed prim paths, then read their affected properties. If the baseline has expired, resetRequired asks for fresh reads. This is a change catalog, not a full scene diff or automatic conflict merge. Files are checked before cached results are reused. Large values carry explicit truncation markers and original lengths or counts; a preview is not a complete replacement value. A write supplies path , revision , a separate output and one action : - edit : author an override layer in .usda or .usdc . edits contains 1–128 operations using op : set attribute , remove sample , visibility , variant or transform . - convert : write .usda , .usdc or .usdz with native USD composition preserved by default. flatten: true explicitly resolves composition for a derived layer or package. - to glb : convert supported USD scene data to one embedded .glb . - from gltf : convert a GLB or glTF into USD text, binary or a package, including supported rig and animation data. - extract : save the subtree at prim as a separate .usda , .usdc , .usdz or .glb asset. Keeps ancestor transforms and required materials, textures and rigs, resolves current composition and variations, and rejects unsupported external object dependencies. format must match the output suffix. GLB requires geometry; USD also accepts material-only selections. For example, this authors a camera-lens override in a new layer: set attribute accepts an optional time and requires a USD type when creating a new property. remove sample needs the property's time code. variant needs an existing set and selection ; an empty selection blocks the weaker layer's variant selection. transform accepts a local correction using translation , quaternion or scale ; it preserves the original transform-operation stack. The write receipt contains output path, outputRevision , byte count, source revision and the conversion report. To replace a different existing output, supply its current outputRevision . Source dependencies cannot be overwritten. Use ?receipt=true to omit file events from an HTTP response; MCP receipts already omit them. Warnings and structured unsupported entries describe conversion limits. A new file does not imply that every USD feature is represented in GLB. Files are limited to 64 MiB each, with 256 MiB and 2,048 files across the dependency graph. Reads and writes also enforce bounded pages, edit batches and conversion budgets. Local source changes invalidate the graph revision; stale writes return 409 .","u":"/docs/api#usd-scenes-in-development"},{"k":"api","t":"API","h":"Runs","x":"POST /runs starts a run and returns { \"runId\" } : Field Meaning --- --- engine claude-code or codex (their own sign-in on this computer), or pi (the Graspable agent) provider For pi : { \"provider\": \"anthropic\", \"modelId\": \"…\", \"apiKey\": \"…\" } . For the others, optionally { \"provider\": \"<engine \", \"modelId\": \"…\" } approvalPolicy auto , guarded (default) or strict budget { \"maxCostUsd\", \"maxTokens\" ","s":"POST /runs starts a run and returns { \"runId\" } : Field Meaning --- --- engine claude-code or codex (their own sign-in on this computer), or pi (the Graspable agent) provider For pi : { \"provider\": \"anthropic\", \"modelId\": \"…\", \"apiKey\": \"…\" } . For the others, optionally { \"provider\": \"<engine \", \"modelId\": \"…\" } approvalPolicy auto , guarded (default) or strict budget { \"maxCostUsd\", \"maxTokens\" } : the run stops when either is passed plugins Secret fields of switched-on plugins: { \"<plugin \": { \"<FIELD \": \"…\" } } Request Does --- --- GET /runs?projectId=:id Runs of a project GET /runs/:id { \"state\", \"events\" } GET /runs/:id/events?since=<seq The same events as a stream (server-sent events) until the run is finished POST /runs/:id/approve { \"toolCallId\", \"approved\" } Answer an approval request POST /runs/:id/steer { \"text\" } Send the agent a message while it works POST /runs/:id/cancel Stop the run POST /runs/:id/revert Undo the run's file changes A run's state is running , waiting user , completed , failed or cancelled .","u":"/docs/api#runs"},{"k":"api","t":"API","h":"Events","x":"Each event is { \"runId\", \"seq\", \"ts\", \"event\" } . The kinds you will use most: event.type Meaning --- --- text The agent said something tool call , tool result A step and its result approval required The run waits for an answer. Send toolCallId to /approve fragment update A file was written: filePath , code verification A check finished: kind ( build , smoke , xr ), pass , summary completion The e","s":"Each event is { \"runId\", \"seq\", \"ts\", \"event\" } . The kinds you will use most: event.type Meaning --- --- text The agent said something tool call , tool result A step and its result approval required The run waits for an answer. Send toolCallId to /approve fragment update A file was written: filePath , code verification A check finished: kind ( build , smoke , xr ), pass , summary completion The end: status ( success , failure , cancelled ) and summary","u":"/docs/api#events"},{"k":"api","t":"API","h":"An example","x":"","u":"/docs/api#an-example"},{"k":"api","t":"API","h":"Film delivery (in development)","x":"These routes are available from 0.17.2. All paths are project-relative and use the same local API authentication as other project routes. Method and path Purpose --- --- GET /projects/:id/films List delivery manifests for the interface GET /projects/:id/films?agent=true Bounded summaries; optional action=list , limit (1–32) and revision-bound cursor GET /projects/:id/films?agent=true&action=inspec","s":"These routes are available from 0.17.2. All paths are project-relative and use the same local API authentication as other project routes. Method and path Purpose --- --- GET /projects/:id/films List delivery manifests for the interface GET /projects/:id/films?agent=true Bounded summaries; optional action=list , limit (1–32) and revision-bound cursor GET /projects/:id/films?agent=true&action=inspect&path=… Manifest or video revision and byte count, without movie contents GET /projects/:id/films?agent=true&action=status Native spatial and projected-video encoder capabilities POST /projects/:id/films?metadata=… Save raw MP4/WebM bytes and a player; metadata is URL-encoded JSON with title, projection and stereo GET /spatial-video/status Apple MV-HEVC availability GET /projected-video/status Apple projected HEVC availability POST /projects/:id/spatial-video Revision-checked native conversion; append ?receipt=true for a compact agent response A native conversion request contains source , its exact SHA-256 sourceRevision , and a separate output MOV path. An existing output needs outputRevision . Choose profile: \"mv-hevc\" (default) or \"apmp-hevc\" . MV-HEVC additionally requires baselineMm , horizontalFovDegrees and disparityAdjustment . An optional manifest attaches the native master to an existing matching film delivery. Source or manifest conflicts preserve the newer file; inspect the response warnings before relying on an updated player. Film saves and human native conversions are recorded in project history and emit file updates. Binary events contain byte counts/hashes and empty code. Run-owned tools record their changes with the owning run. Writes refuse an active conflicting agent run. Uploads are bounded to 192 MB; native conversion has separate media and duration limits. The existing publish request accepts film (manifest path) and filmRevision . This deploys the selected film player instead of building the interactive project, under the usual account and visibility rules. See Share immersive films for formats, cloud limits and playback workflows.","u":"/docs/api#film-delivery-in-development"},{"k":"docs","t":"Cloud Agents","x":"Run a project in the cloud instead of on your computer, keep it going while you are away, and move it back. A project normally runs on your computer. You can move one to the cloud. The agent, its commands and the preview then run there, and keep running when your computer sleeps. You can move it back at any time. A project lives in one place at a time: This computer or Cloud . The button in the pr","s":"Run a project in the cloud instead of on your computer, keep it going while you are away, and move it back. A project normally runs on your computer. You can move one to the cloud. The agent, its commands and the preview then run there, and keep running when your computer sleeps. You can move it back at any time. A project lives in one place at a time: This computer or Cloud . The button in the project header shows which.","u":"/docs/cloud-agents"},{"k":"docs","t":"Cloud Agents","h":"What you need","x":"- To be signed in under Settings → Graspable account . - Credits. A new account gets 6 once, enough for an hour. Paid plans include Credits every month. - The Graspable agent with an API key account. A subscription sign-in does not work in the cloud; see below. If your default is a subscription and you have also saved an API key, a cloud project uses the API key by itself. The list above the promp","s":"- To be signed in under Settings → Graspable account . - Credits. A new account gets 6 once, enough for an hour. Paid plans include Credits every month. - The Graspable agent with an API key account. A subscription sign-in does not work in the cloud; see below. If your default is a subscription and you have also saved an API key, a cloud project uses the API key by itself. The list above the prompt shows which one, and lets you choose another. There is nothing to set up. Cloud Agents run in Graspable cloud and use Credits: 1 Credit for every 10 minutes a Cloud Agent runs.","u":"/docs/cloud-agents#what-you-need"},{"k":"docs","t":"Cloud Agents","h":"Move a project to the cloud","x":"1. Open the project. Wait for any run to finish. 2. Click This computer in the header. 3. Choose Move to Graspable cloud . The menu shows how many Credits you have left. The project's files, run history and memory travel with it. Only when the cloud has everything is the copy on your computer removed. The header then reads Cloud .","u":"/docs/cloud-agents#move-a-project-to-the-cloud"},{"k":"docs","t":"Cloud Agents","h":"Work on a cloud project","x":"It works like a local project: the editor, runs, approvals, memory and the preview. A Cloud Agent pauses by itself a few minutes after you stop using the project, and wakes when you use it again. A paused Cloud Agent uses no Credits and keeps your files. A run that has started gets up to an hour to finish, even if you close the app. From the header button you can also choose Pause now (stops the c","s":"It works like a local project: the editor, runs, approvals, memory and the preview. A Cloud Agent pauses by itself a few minutes after you stop using the project, and wakes when you use it again. A paused Cloud Agent uses no Credits and keeps your files. A run that has started gets up to an hour to finish, even if you close the app. From the header button you can also choose Pause now (stops the cost; files are kept) or Wake now .","u":"/docs/cloud-agents#work-on-a-cloud-project"},{"k":"docs","t":"Cloud Agents","h":"The first run after a move","x":"The agent does not remember earlier runs by itself. So after a project moves, to the cloud or back, the first run starts with a short account of the earlier work: what was asked, which files changed, how each run ended, and whether you undid it. You see it in the run's Context list as Earlier work on this project . The account can be made in three ways. The app uses the first one that is available","s":"The agent does not remember earlier runs by itself. So after a project moves, to the cloud or back, the first run starts with a short account of the earlier work: what was asked, which files changed, how each run ended, and whether you undid it. You see it in the run's Context list as Earlier work on this project . The account can be made in three ways. The app uses the first one that is available: How When Cost --- --- --- System One rates every passage of the earlier conversation and keeps the ones your new request needs You are signed in and System One is in Assist mode (the default) Included Fast mode cuts the conversation down Fast mode is on 0.1 to 0.2 Credits A short list: each request, the files changed and the result Always Nothing","u":"/docs/cloud-agents#the-first-run-after-a-move"},{"k":"docs","t":"Cloud Agents","h":"After you update the app","x":"A cloud project keeps working when you update Graspable: the app and the Cloud Agent do not have to be the same version. The Cloud Agent itself stays on the version it was started with. To get the newer agent, click Cloud in the project header and choose Update the cloud agent . Graspable starts a new Cloud Agent, moves the project to it and removes the old one. In Graspable cloud this uses 1 Cred","s":"A cloud project keeps working when you update Graspable: the app and the Cloud Agent do not have to be the same version. The Cloud Agent itself stays on the version it was started with. To get the newer agent, click Cloud in the project header and choose Update the cloud agent . Graspable starts a new Cloud Agent, moves the project to it and removes the old one. In Graspable cloud this uses 1 Credit and takes a minute or two. In a team, a teammate who had the project open chooses Open again under Shared with your team.","u":"/docs/cloud-agents#after-you-update-the-app"},{"k":"docs","t":"Cloud Agents","h":"Work on it with other people","x":"On the Studio plan you can share a cloud project with your team. Everyone in the team can then open it and work on the same copy. See Teams.","u":"/docs/cloud-agents#work-on-it-with-other-people"},{"k":"docs","t":"Cloud Agents","h":"Move it back","x":"Click the Cloud button and choose Move back to this computer . The project is copied back and then deleted from the cloud. Moving back is free. You cannot move a project while a run is going. Editing pauses during the move. If files change outside Graspable before the move finishes, the newer copy is kept and you can retry. On current Cloud Agents, moving back reuses unchanged files from the backu","s":"Click the Cloud button and choose Move back to this computer . The project is copied back and then deleted from the cloud. Moving back is free. You cannot move a project while a run is going. Editing pauses during the move. If files change outside Graspable before the move finishes, the newer copy is kept and you can retry. On current Cloud Agents, moving back reuses unchanged files from the backup already on your computer. Large models and textures do not need to download again. Files are checked by their contents before reuse. A fresh cloud sandbox still needs the initial upload. When you use your own server, recent transfers can also reuse unchanged files on the next upload. The server keeps a private temporary cache for up to 30 minutes, with a combined limit of 512 MiB across four projects. Larger projects still move, but are not cached. Older Cloud Agents continue to use a full transfer; update the agent to get these improvements.","u":"/docs/cloud-agents#move-it-back"},{"k":"docs","t":"Cloud Agents","h":"What does not work in the cloud","x":"Not available Why --- --- A ChatGPT or Claude subscription sign-in A subscription sign-in stays on your computer Claude Code and the Codex CLI They use the sign-in on your computer A model on your computer or local network A cloud agent cannot reach it Headset preview Not available for cloud projects yet. Move the project back first So a cloud run needs the Graspable agent with an API key account.","s":"Not available Why --- --- A ChatGPT or Claude subscription sign-in A subscription sign-in stays on your computer Claude Code and the Codex CLI They use the sign-in on your computer A model on your computer or local network A cloud agent cannot reach it Headset preview Not available for cloud projects yet. Move the project back first So a cloud run needs the Graspable agent with an API key account. If you try something else, the run is refused with an explanation. Your API key is sent with each run and is not stored in the cloud.","u":"/docs/cloud-agents#what-does-not-work-in-the-cloud"},{"k":"docs","t":"Cloud Agents","h":"How much fits","x":"A cloud project may be so large, and all your cloud projects together so large, depending on your plan. The size is the project's own files as they are sent up, without its packages. Code is small; splats and scanned rooms are what counts. Plan One project All cloud projects together --- --- --- Explorer 100 MB 200 MB Builder 250 MB 1 GB Pro 500 MB 3 GB Studio 1 GB 10 GB A project that does not fi","s":"A cloud project may be so large, and all your cloud projects together so large, depending on your plan. The size is the project's own files as they are sent up, without its packages. Code is small; splats and scanned rooms are what counts. Plan One project All cloud projects together --- --- --- Explorer 100 MB 200 MB Builder 250 MB 1 GB Pro 500 MB 3 GB Studio 1 GB 10 GB A project that does not fit is not moved: Graspable says how large it is and what your plan allows, and nothing is started or charged. To make room, move another project back to this computer. A .ply splat saved as .spz is about ten times smaller. On a team, the room is the owner's and is shared.","u":"/docs/cloud-agents#how-much-fits"},{"k":"docs","t":"Cloud Agents","h":"Costs","x":"In Graspable cloud a Cloud Agent uses 1 Credit for every 10 minutes it runs. Your plan includes Credits each month; Credits explains the details. Your AI provider bills your API key as usual.","u":"/docs/cloud-agents#costs"},{"k":"docs","t":"Cloud Agents","h":"The backup","x":"When you move a project up, the copy that was sent is kept on your computer in ~/Graspable/projects/.cloud . If the sandbox is lost, the header button shows \"not found\" and offers Restore the copy from before the move . Work done in the cloud after the move is not in that copy. The backup is removed after the project has moved back.","u":"/docs/cloud-agents#the-backup"},{"k":"docs","t":"Cloud Agents","h":"Your own cloud","x":"You can run Cloud Agents in your own cloud instead: your own E2B account (Builder and higher), or a server you run (Pro and higher). These use no Credits; your provider bills you. Open Settings → Cloud Agents . E2B sandbox. Enter your E2B API key and choose Save . The Domain field is only for E2B running in your own cloud. Your E2B account must hold an agent image for your exact Graspable version;","s":"You can run Cloud Agents in your own cloud instead: your own E2B account (Builder and higher), or a server you run (Pro and higher). These use no Credits; your provider bills you. Open Settings → Cloud Agents . E2B sandbox. Enter your E2B API key and choose Save . The Domain field is only for E2B running in your own cloud. Your E2B account must hold an agent image for your exact Graspable version; ask us for it when you set this up. Your own server. Enter the Agent address and the Agent token from your admin, and choose Save . The address must start with https:// , and the agent on the server must be the same Graspable version as your app. These details are stored in your computer's keychain. The project header then also offers Move to E2B sandbox or Move to Your own server . Setting up a server needs our help: use Contact sales on the pricing page.","u":"/docs/cloud-agents#your-own-cloud"},{"k":"docs","t":"Teams","x":"Work with other people on the same cloud projects, with one plan, one pool of Credits and shared project memory. A team is a group of people on one Studio plan. Everyone in the team: - uses the Studio plan, - spends from one shared pool of Credits, - can open the cloud projects shared with the team. Studio is priced per seat and starts at 3 seats. A seat is one person, or one invitation that is st","s":"Work with other people on the same cloud projects, with one plan, one pool of Credits and shared project memory. A team is a group of people on one Studio plan. Everyone in the team: - uses the Studio plan, - spends from one shared pool of Credits, - can open the cloud projects shared with the team. Studio is priced per seat and starts at 3 seats. A seat is one person, or one invitation that is still open. See Plans and billing.","u":"/docs/teams"},{"k":"docs","t":"Teams","h":"Start a team","x":"1. Subscribe to Studio in Settings → Plan and Credits and choose the number of seats. 2. Open Settings → Team . Your team is there, with you as its owner.","u":"/docs/teams#start-a-team"},{"k":"docs","t":"Teams","h":"Invite someone","x":"1. In Settings → Team , enter their email and choose Invite . 2. They open Graspable, sign in with that email under Settings → Graspable account , and open Settings → Team . 3. They choose Join . They get an email that says how to join. An invitation is valid for 14 days and only for that email address. If all seats are in use, change Seats first. The difference is charged or credited on the next ","s":"1. In Settings → Team , enter their email and choose Invite . 2. They open Graspable, sign in with that email under Settings → Graspable account , and open Settings → Team . 3. They choose Join . They get an email that says how to join. An invitation is valid for 14 days and only for that email address. If all seats are in use, change Seats first. The difference is charged or credited on the next invoice.","u":"/docs/teams#invite-someone"},{"k":"docs","t":"Teams","h":"Roles","x":"Role Can do --- --- Owner Everything, including billing, seats, Credit packs and overage Admin Invite and remove people; move any team project back from the cloud Member Use the plan, the Credits and the shared projects The owner changes a role in the list of people.","u":"/docs/teams#roles"},{"k":"docs","t":"Teams","h":"Share a project","x":"A shared project is a Cloud Agent project in Graspable cloud. There is one copy, in the cloud, and everyone works on it. 1. Open the project and click This computer in the header. 2. Choose Move to Graspable cloud and share with your team. To share a project that is already in Graspable cloud, click Cloud in the header and choose Share with your team.","u":"/docs/teams#share-a-project"},{"k":"docs","t":"Teams","h":"Open a shared project","x":"On the project list, shared projects appear under Shared with your team. Choose Open . Nothing is copied to your computer; your app connects to the project in the cloud. What is shared: Shared Not shared --- --- Files, run history and the preview Your AI account. Each person runs the agent with their own API key The project's memory Your personal memory across projects The cost: Credits come from ","s":"On the project list, shared projects appear under Shared with your team. Choose Open . Nothing is copied to your computer; your app connects to the project in the cloud. What is shared: Shared Not shared --- --- Files, run history and the preview Your AI account. Each person runs the agent with their own API key The project's memory Your personal memory across projects The cost: Credits come from the team Only one run can go at a time in a project. If a teammate has a run going, you are asked to wait until it is finished.","u":"/docs/teams#open-a-shared-project"},{"k":"docs","t":"Teams","h":"Who used what","x":"Settings → Team shows the Credits each person used this month.","u":"/docs/teams#who-used-what"},{"k":"docs","t":"Teams","h":"Move a shared project back","x":"The person who moved it to the cloud, or a team owner or admin, can choose Move back to this computer . The project then lives on that person's computer and is no longer shared. A member who opened a shared project can choose Remove from my list . The project stays with the team.","u":"/docs/teams#move-a-shared-project-back"},{"k":"docs","t":"Teams","h":"Leave or remove someone","x":"In Settings → Team , choose Leave on your own row, or Remove on someone else's. That person is back on their own plan and can no longer wake or open team projects. A project they had already opened stays reachable from their app until it is paused or moved. To be sure, move the project back to your computer and share it again.","u":"/docs/teams#leave-or-remove-someone"},{"k":"docs","t":"Credits","x":"What Credits pay for, how many each plan includes, Credit packs, overage, and what happens when they run out. Credits pay for things Graspable runs for you. Building on your own computer never uses Credits.","u":"/docs/credits"},{"k":"docs","t":"Credits","h":"What uses Credits","x":"What Cost --- --- A Cloud Agent in Graspable cloud 1 Credit for every 10 minutes it runs A Fast mode edit, shorter output or progress check 0.1 to 0.3 Credits; more on very large files A Fast mode code search About 1 Credit System One (fast decisions) Included. Beyond about 400 decisions in a day: 0.1 Credit for about every 130 more Voice with your Graspable account About ten minutes of speech eve","s":"What Cost --- --- A Cloud Agent in Graspable cloud 1 Credit for every 10 minutes it runs A Fast mode edit, shorter output or progress check 0.1 to 0.3 Credits; more on very large files A Fast mode code search About 1 Credit System One (fast decisions) Included. Beyond about 400 decisions in a day: 0.1 Credit for about every 130 more Voice with your Graspable account About ten minutes of speech every day are free. Beyond them: about a third of a Credit for ten more minutes of speaking, and about a Credit for a page of answers read aloud Voice with your own ElevenLabs key Nothing Building on this computer Nothing The AI model, with your own AI account Nothing The AI model, with the Graspable agent on Credits By what each request used: 10 Credits for every dollar the model costs A call your published app makes to one of your connected services 0.2 Credits A Cloud Agent is paid for in blocks of 10 minutes. While you work on a cloud project, Graspable keeps it paid a little ahead. When you start a run, it pays up to an hour ahead so the run can finish while you are away. When the paid time ends and you are not using the project, the Cloud Agent pauses. A paused Cloud Agent uses no Credits and keeps your files. Time that was paid ahead and not used is not returned. Fast mode comes with the Pro plan and higher. A step that fails costs nothing.","u":"/docs/credits#what-uses-credits"},{"k":"docs","t":"Credits","h":"How many you get","x":"Plan Credits About this much Cloud Agent time --- --- --- Explorer 6, once 1 hour, to try a Cloud Agent Builder 150 a month 25 hours Pro 250 a month 42 hours Studio 500 a month per seat, shared by the team 83 hours per seat For the first 100 accounts to take Builder or Pro, the plan gives three times these Credits each month for the first three months. The pricing page says how many places are lef","s":"Plan Credits About this much Cloud Agent time --- --- --- Explorer 6, once 1 hour, to try a Cloud Agent Builder 150 a month 25 hours Pro 250 a month 42 hours Studio 500 a month per seat, shared by the team 83 hours per seat For the first 100 accounts to take Builder or Pro, the plan gives three times these Credits each month for the first three months. The pricing page says how many places are left. The offer starts when you take the plan and does not apply to Studio or to Credit packs. Credits renew each month with your plan. Unused plan Credits do not carry over. The 6 Explorer Credits are given once and last 30 days.","u":"/docs/credits#how-many-you-get"},{"k":"docs","t":"Credits","h":"Generated worlds","x":"Asking for a new place in a world template uses Credits: 2 for a draft, 11 for a standard world, 21 for a large one. A world that fails to generate costs nothing. With your own World Labs key, worlds use no Credits. See Generated worlds.","u":"/docs/credits#generated-worlds"},{"k":"docs","t":"Credits","h":"Credit packs","x":"A Credit pack is a one-time purchase, for any account. Pack Price --- --- 100 Credits $20 300 Credits $54 1,000 Credits $150 You can also buy any amount from 50 Credits. The price per Credit is $0.20, $0.18 from 300 Credits and $0.15 from 1,000 Credits. Pack Credits are used after your plan's Credits for the month. They last 12 months and are not lost when the month ends or when your plan changes.","s":"A Credit pack is a one-time purchase, for any account. Pack Price --- --- 100 Credits $20 300 Credits $54 1,000 Credits $150 You can also buy any amount from 50 Credits. The price per Credit is $0.20, $0.18 from 300 Credits and $0.15 from 1,000 Credits. Pack Credits are used after your plan's Credits for the month. They last 12 months and are not lost when the month ends or when your plan changes. Buy Credits in Settings → Plan and Credits , under Buy Credits : choose a pack, or type an amount next to Another amount and choose Buy . In a team, the owner buys packs for the team. Where a pack shows a monthly price, you can also have it every month : the same pack is bought again each month until you stop it under Payment method and billing details . Each month's Credits last 12 months like any pack. Pay with crypto (experimental) You can pay for Credits with crypto from your own wallet instead of a card. Prices are the same, and there is no identity check: you need your Graspable account and a wallet that holds the coin. 1. In Settings → Plan and Credits , under Buy Credits , choose Pay with crypto . 2. Choose how many Credits, then a coin. The app shows the coins that can be taken right now. 3. The app shows the amount, an address and a QR code. Pay from your wallet: scan the code, or copy the address and the amount. 4. The Credits are added once the network has confirmed the payment. A payment that is only announced counts for nothing yet; the app shows how many confirmations there are and how many are needed. USDC on Base and Solana take under a minute, Ether a few minutes. Bitcoin needs one confirmation for a purchase up to $100 (about 10 minutes, at times longer), three up to $1,000 and six above. You can close the app meanwhile: the order is shown again when you come back, in the list of your purchases. On the account website, copy buttons confirm success after your browser accepts the value. If clipboard access is blocked, the full address, invoice or amount appears in a selectable field for manual copying. What to know: - The price holds for 20 minutes. A payment that arrives later the same day is still credited at the price shown. - When the app says \"Send exactly\", send the exact amount. For Ether, Solana and USDC, and at times for Bitcoin, every payment goes to the same address, and the last digits of the amount are how yours is told apart. Send it on the network shown; a payment on another network does not arrive. - If less arrives than asked (Bitcoin to an address of its own), you get the Credits that amount pays for at the price shown. - A payment cannot be taken back. If you paid and the Credits did not arrive within a day, write to us with the order reference the app shows. Graspable holds no wallet for this and never asks for yours. Agents never buy Credits. In a team, the owner buys them.","u":"/docs/credits#credit-packs"},{"k":"docs","t":"Credits","h":"Overage","x":"Overage lets work continue when your Credits run out. It is off until you turn it on, and it is for paid plans. 1. In Settings → Plan and Credits , tick Keep going when Credits run out . 2. Set the most extra Credits you allow in a month. Extra Credits are added to your next invoice at your plan's rate: $0.20 on Builder, $0.18 on Pro, $0.15 on Studio. Graspable never goes past the limit you set.","u":"/docs/credits#overage"},{"k":"docs","t":"Credits","h":"The order Credits are used in","x":"1. This month's plan Credits. 2. Credit packs, the one that expires first. 3. Overage, if you turned it on, up to your limit.","u":"/docs/credits#the-order-credits-are-used-in"},{"k":"docs","t":"Credits","h":"See what you have left","x":"Your Credits are shown at the top of the Projects page and of every project; click the number to open Settings → Plan and Credits , which shows how many are left of this month's plan Credits, when new ones come, and how many you bought. The number turns amber when you are running low. The This computer button in a project also shows them before you move the project to the cloud.","u":"/docs/credits#see-what-you-have-left"},{"k":"docs","t":"Credits","h":"When you run out","x":"Your Cloud Agents pause when their paid time ends, and runs carry on without Fast mode. Nothing is lost. You can: - buy a Credit pack or turn on overage, - wait for your Credits to renew, - move to a higher plan in Settings → Plan and Credits , or - move the project back to your computer, which is free.","u":"/docs/credits#when-you-run-out"},{"k":"docs","t":"Credits","h":"In a team","x":"A team has one pool of Credits: 500 a month for every seat. Everyone in the team spends from it. Settings → Team shows what each person used this month. See Teams.","u":"/docs/credits#in-a-team"},{"k":"docs","t":"Credits","h":"Your own cloud","x":"On the Studio and Enterprise plans you can run Cloud Agents in your own cloud instead. Those use no Credits: your cloud provider bills you. See Cloud Agents.","u":"/docs/credits#your-own-cloud"},{"k":"docs","t":"Plans and billing","x":"What is free, what each plan adds, and how to subscribe, change or cancel. Building on your own computer with your own AI account is free, on every plan and without limits. A plan is for what Graspable runs for you.","u":"/docs/plans-and-billing"},{"k":"docs","t":"Plans and billing","h":"The plans","x":"Explorer Builder Pro Studio --- --- --- --- --- Price a month Free $20 $40 $99 per seat Price a month, billed yearly Free $16 $32 $79 per seat Local projects with your own AI account Yes Yes Yes Yes Community templates Yes Yes Yes Yes Premium templates No Yes Yes Yes Native XR: Unity, Unreal Engine and Apple Vision Pro No No Yes Yes Headset app as a download with a published native app No No Up to","s":"Explorer Builder Pro Studio --- --- --- --- --- Price a month Free $20 $40 $99 per seat Price a month, billed yearly Free $16 $32 $79 per seat Local projects with your own AI account Yes Yes Yes Yes Community templates Yes Yes Yes Yes Premium templates No Yes Yes Yes Native XR: Unity, Unreal Engine and Apple Vision Pro No No Yes Yes Headset app as a download with a published native app No No Up to 500 MB Up to 1 GB Credits a month 6, once 150 250 500 per seat Cloud Agents in Graspable cloud With Credits Yes Yes Yes Voice, with free minutes every day and Credits beyond them Yes Yes Yes Yes Credit packs Yes Yes Yes Yes Overage, per extra Credit No $0.20 $0.18 $0.15 Fast mode No No Yes Yes Team: shared projects and Credits No No No Yes Cloud Agents in your own E2B account No Yes Yes Yes Cloud Agents on your own server No No Yes Yes Yearly billing is charged once a year: $192 for Builder, $384 for Pro and $948 per seat for Studio. Studio starts at 3 seats. Need more, or a custom agreement? The pricing page has a Contact sales link for an Enterprise plan.","u":"/docs/plans-and-billing#the-plans"},{"k":"docs","t":"Plans and billing","h":"What you still pay others","x":"- Your AI provider, for your subscription or your API key use. Credits pay for the model only when you choose the Graspable agent on Credits. - Your cloud provider, if you run Cloud Agents in your own cloud. - ElevenLabs, if you use voice with your own ElevenLabs key.","u":"/docs/plans-and-billing#what-you-still-pay-others"},{"k":"docs","t":"Plans and billing","h":"See your plan","x":"1. Sign in under Settings → Graspable account . 2. Open Settings → Plan and Credits . You see your plan, your Credits and when they renew. The same is on the website: sign in and open your account. There you can subscribe, buy Credits, see payments and invoices, manage your team and your published apps, without the app.","u":"/docs/plans-and-billing#see-your-plan"},{"k":"docs","t":"Plans and billing","h":"Subscribe or upgrade","x":"1. In Settings → Plan and Credits , choose Monthly or Yearly . 2. For Studio, set the number of Seats . 3. Choose Subscribe or Upgrade on a plan. 4. Checkout opens in your browser. Payment is handled by Polar.","u":"/docs/plans-and-billing#subscribe-or-upgrade"},{"k":"docs","t":"Plans and billing","h":"Pay for a plan with crypto (experimental)","x":"You can pay for Builder, Pro or Studio with crypto from your own wallet instead of a card. There is no card, no name and no identity check: you need your Graspable account and a wallet that holds the coin. A wallet only ever sends money, so nothing can be taken from it each month. You pay for time ahead: 1 month, 3 months or 1 year. The prices are the card's: a month at the monthly price, a year a","s":"You can pay for Builder, Pro or Studio with crypto from your own wallet instead of a card. There is no card, no name and no identity check: you need your Graspable account and a wallet that holds the coin. A wallet only ever sends money, so nothing can be taken from it each month. You pay for time ahead: 1 month, 3 months or 1 year. The prices are the card's: a month at the monthly price, a year at the yearly price. 1. In the app, open Settings → Plan and Credits , choose Pay with crypto , then A plan, paid ahead . On the website, sign in and open your account. 2. Choose the plan and how long. For Studio, set the number of seats. 3. Choose a coin. You see the amount, an address and a QR code. Pay from your wallet. 4. The plan starts once the network has confirmed the payment. Paying works as it does for Credits: the price holds for 20 minutes, and where it says \"Send exactly\", send the exact amount. What to know: - It does not renew. The plan runs until the day shown. We send an email a week before and a day before. If you do nothing, your account goes back to Explorer on that day. Your projects on your computer and your published apps stay. - More time is added to the end. Paying early loses nothing. - A dearer plan, or another seat, while time is left: you pay the difference for the time you have already paid. Choose No more time, only the change if you want nothing else. - A cheaper plan, or fewer seats: what is left is turned into more days of the cheaper plan. - Credits are the plan's own, every 30 days while the plan runs. Extra Credits beyond the plan (overage) need a card; with crypto, buy a Credit pack instead. - If less than the price arrives, the plan does not start, and the money becomes Credits at their usual price. - Card and crypto do not mix. With a plan paid by card, cancel it first and pay with crypto when its time has ended. - A payment cannot be taken back.","u":"/docs/plans-and-billing#pay-for-a-plan-with-crypto-experimental"},{"k":"docs","t":"Plans and billing","h":"More Credits","x":"Both are in Settings → Plan and Credits . Credits explains them. - Buy Credits. Choose a pack, or type another amount and choose Buy . Checkout opens in your browser, and the Credits are added when the payment is done. - Keep going when Credits run out (overage). Tick Allow extra Credits and set the Monthly limit . Extra Credits are added to your next invoice.","u":"/docs/plans-and-billing#more-credits"},{"k":"docs","t":"Plans and billing","h":"Cancel, resume or change payment details","x":"In Settings → Plan and Credits : - Cancel at period end stops the renewal. You keep the plan until the period ends. - Resume undoes a cancellation before the period ends. - Manage billing opens the billing portal in your browser, for invoices and payment details. After a plan ends you are on Explorer. Your local projects keep working. A project that is in Graspable cloud stays there, paused; move ","s":"In Settings → Plan and Credits : - Cancel at period end stops the renewal. You keep the plan until the period ends. - Resume undoes a cancellation before the period ends. - Manage billing opens the billing portal in your browser, for invoices and payment details. After a plan ends you are on Explorer. Your local projects keep working. A project that is in Graspable cloud stays there, paused; move it back to your computer when you like. Credit packs you bought stay.","u":"/docs/plans-and-billing#cancel-resume-or-change-payment-details"},{"k":"docs","t":"Plans and billing","h":"In a team","x":"If you are a member of a team, your plan and Credits come from the team. Only the team owner sees and changes billing. See Teams.","u":"/docs/plans-and-billing#in-a-team"},{"k":"docs","t":"Privacy and telemetry","x":"Exactly what Graspable sends, what it never sends, how to turn it off, and where your keys and sign-ins are stored. ","u":"/docs/privacy-and-telemetry"},{"k":"docs","t":"Privacy and telemetry","h":"What stays on your computer","x":"By default, everything about a project is on your computer: its files, the agent, the commands it runs, the preview, the checks, run history and memory. Two things leave your computer because they must: - Your requests and the code the agent reads go to the model provider you chose (OpenAI, Anthropic, or another), so the model can answer. That provider's terms apply. With a local model, they stay ","s":"By default, everything about a project is on your computer: its files, the agent, the commands it runs, the preview, the checks, run history and memory. Two things leave your computer because they must: - Your requests and the code the agent reads go to the model provider you chose (OpenAI, Anthropic, or another), so the model can answer. That provider's terms apply. With a local model, they stay on your computer. - A cloud project runs in Graspable cloud (sandboxes at our provider, E2B), or in your own cloud if you set one up. See Cloud Agents. - Fast mode sends the step it speeds up (the file being edited, the lines a search reads, or the command output) to Graspable and our provider for that step, Morph. Graspable does not store it. Fast mode is off unless you turn it on. See Fast mode. - Voice sends what you say to ElevenLabs while you use the microphone beside the prompt: directly with your own ElevenLabs key, or through Graspable on Credits. Graspable keeps no recording and no text of it, only how many seconds were turned into text that day. See Voice. - The app asks graspable.dev for the current agent instructions. The texts Graspable writes for the agent can be updated without an app update. The request carries nothing about you or your project, and the app works offline with the texts it last saw. Graspable itself receives only what this page lists.","u":"/docs/privacy-and-telemetry#what-stays-on-your-computer"},{"k":"docs","t":"Privacy and telemetry","h":"Run summaries","x":"After each finished run, the app sends one anonymized summary to Graspable. This is on by default, and it is sent only while you are signed in to a Graspable account. The summary contains: - the engine, model and template, - the version of Graspable's own instructions to the agent, and the text of those instructions (written by Graspable, the same for everyone), - the outcome and duration, - count","s":"After each finished run, the app sends one anonymized summary to Graspable. This is on by default, and it is sent only while you are signed in to a Graspable account. The summary contains: - the engine, model and template, - the version of Graspable's own instructions to the agent, and the text of those instructions (written by Graspable, the same for everyone), - the outcome and duration, - counts: events, tool calls, approvals, denials, repairs, memory items recalled and saved, - whether the build passed first time, - tokens and cost, - System One decisions, as numbers and labels, - the app version, - if the team is testing a variant of the agent's instructions, which variant your run got. You can see that variant in the Context list of the run. The summary is tied to a run key made by scrambling your account ID with the run ID, not to your name.","u":"/docs/privacy-and-telemetry#run-summaries"},{"k":"docs","t":"Privacy and telemetry","h":"Error reports","x":"When something breaks in Graspable itself, the app sends a short error report so it can be fixed: an error in the app's window, in the agent on your computer, or a run that ended with an error. A report holds: - the kind of error and its message, - for errors in the app or the agent, where in Graspable's own code it happened, - the version of Graspable and your system (macOS, Windows or Linux). Be","s":"When something breaks in Graspable itself, the app sends a short error report so it can be fixed: an error in the app's window, in the agent on your computer, or a run that ended with an error. A report holds: - the kind of error and its message, - for errors in the app or the agent, where in Graspable's own code it happened, - the version of Graspable and your system (macOS, Windows or Linux). Before a report leaves your computer, the path of your home folder, email addresses and anything that looks like a key or a token are taken out of it, and the message is cut to a few hundred characters. Graspable's server does the same again and keeps nothing about who sent it: a report is not tied to your account. Upstream HTTP error bodies are omitted from reports, including any balance or request text they contain. The HTTP status is kept, including when a provider omits the HTTP prefix. Multiline response bodies are removed from stack reports too. Sentry is told not to infer an IP address for a desktop report. A report never contains your requests, your code, or what the agent did.","u":"/docs/privacy-and-telemetry#error-reports"},{"k":"docs","t":"Privacy and telemetry","h":"Usage of the app","x":"The app counts how it is used, so we can see what people do with it and what they never find. It sends these events, and only these: - the app was opened, and which screen is open (the Projects page, a project, Settings), without the project's name or id, - you signed in to your Graspable account, added an AI account (which provider), or started signing in to an agent, - a project was created and ","s":"The app counts how it is used, so we can see what people do with it and what they never find. It sends these events, and only these: - the app was opened, and which screen is open (the Projects page, a project, Settings), without the project's name or id, - you signed in to your Graspable account, added an AI account (which provider), or started signing in to an agent, - a project was created and from which template; a premium template was chosen without the plan for it; a project was made from a published app; a capture was kept or chosen as a project's picture (never the picture itself), - a run started (which agent, which template, on your computer or in the cloud, how many connected services) and how it ended, in numbers: how long it took, how many steps, repairs, approvals and tokens, its cost at your AI provider, whether it built first time, - a scene was shipped (a run that ended with a project that builds and loads), and whether it was your first, - you answered an approval, undid a run, restored a version, or rated a message 👍 or 👎, - the preview was opened, the project was shared to a headset (Wi-Fi or USB), runs were compared, a world was added, - voice was used, and how (a spoken request, a conversation, talking in the headset, or a message read aloud), never what was said, - an app was published, as public or private; a project was moved to the cloud or back; a project was linked to GitHub, - a connected service (which one), a plugin or a skill was added, - the page for buying a plan or Credits was opened. A project appears in these events as a short code made from its id, so shipped scenes can be counted per project. The code cannot be turned back into the project's name. Each event carries the app's version and your system. Once you are signed in to your Graspable account it also carries your account id, never your email, so a download can be followed to a first project. The app does not record what is on the screen: no clicks, no typed text, no session recordings. A request, your code, a file name or a project's name cannot leave this way. These events go to PostHog, the analytics service Graspable uses, which stores them in the United States. Graspable's own server also records in PostHog what it does for your account: that a plan started, changed or ended, that Credits were bought or spent and on what, that a Cloud Agent started, and how many times a published app was visited. These carry your account id and amounts, never a card or an email. The same counts of runs, added up over everyone, make the public model leaderboard: how often each model finishes, what a run costs and how long it takes. A model is shown only after five runs by at least three people, so no person's use can be read from it. Emails. A new account gets one email a day for its first week, from Michal, about getting started. The welcome arrives first; the following emails are scheduled for 10:00 in your time zone, including across daylight-saving changes. If your time zone is unavailable, they use the time of day you signed up. Every one has a link that stops them at once, and none is sent after that. Emails about your own account, such as a sign-in link, are separate and always sent. The first-week emails record whether each was delivered and opened, and whether a link in it was followed, so we can tell which of them are worth sending; account emails are not tracked this way. What you do in Community is kept with your account, because it is what the feature is: which apps you liked, whom you follow, and your collections. Likes are shown as a count, never as a list of names; a public collection and your follower count are shown on your handle's page.","u":"/docs/privacy-and-telemetry#usage-of-the-app"},{"k":"docs","t":"Privacy and telemetry","h":"Use from grasp, MCP and the API","x":"When Graspable is used from outside the app, by you in a terminal or by another AI agent, it counts that use so we can see how agents work with it: - a grasp command (its name only, such as run or publish ), an MCP tool call (the tool's name), or a call to the local API made with an access token (its route, such as POST /projects/:id/preview ), - the client that made it (an MCP client's name, such","s":"When Graspable is used from outside the app, by you in a terminal or by another AI agent, it counts that use so we can see how agents work with it: - a grasp command (its name only, such as run or publish ), an MCP tool call (the tool's name), or a call to the local API made with an access token (its route, such as POST /projects/:id/preview ), - the client that made it (an MCP client's name, such as Claude Code or Cursor, or the program's name for an API call), - which agent started grasp , when one can be told from its environment (Claude Code, Codex, Cursor, a terminal, a script, CI), - whether it worked and how long it took, - the version, your system, and an installation category inferred from the program's location (Homebrew, the install script, the app, development or unknown). The location itself is never sent. They carry a random id made on your computer, not your account, and never the arguments: no requests, paths, file names or output. To turn this off, run grasp telemetry off , untick the switch in the app (the two share it), or set DO NOT TRACK=1 or GRASPABLE TELEMETRY=0 .","u":"/docs/privacy-and-telemetry#use-from-grasp-mcp-and-the-api"},{"k":"docs","t":"Privacy and telemetry","h":"Welcome questions","x":"A little while after the app starts, it asks four optional questions in a dialog: how you heard about Graspable, what you want to make, what describes you, and which headsets you have. Each takes several answers. Choose Not now to close it: it then comes up once each time the app starts, and a button with a red dot in the header opens it whenever you like, until you answer. Your answers are kept w","s":"A little while after the app starts, it asks four optional questions in a dialog: how you heard about Graspable, what you want to make, what describes you, and which headsets you have. Each takes several answers. Choose Not now to close it: it then comes up once each time the app starts, and a button with a red dot in the header opens it whenever you like, until you answer. Your answers are kept with your Graspable account, so you are asked once whichever computer you use. If you are not signed in, they stay on this computer and are sent when you sign in. If you pick \"Other\" for how you heard about us, the few words you type are kept too. While Help improve Graspable is on, the same answers also go with your usage events.","u":"/docs/privacy-and-telemetry#welcome-questions"},{"k":"docs","t":"Privacy and telemetry","h":"The download link by email","x":"On a phone, the website offers to email you the download link, because Graspable installs on a computer. If you use it, we keep your address, the tag of the link that brought you to the site (for example which post or ad), and whether you opened the email. We send that one email. If you ticked the box, three short tips from Michal follow over the next week, and news when something new ships; nothi","s":"On a phone, the website offers to email you the download link, because Graspable installs on a computer. If you use it, we keep your address, the tag of the link that brought you to the site (for example which post or ad), and whether you opened the email. We send that one email. If you ticked the box, three short tips from Michal follow over the next week, and news when something new ships; nothing else. Every email has a link that makes us forget your address.","u":"/docs/privacy-and-telemetry#the-download-link-by-email"},{"k":"docs","t":"Privacy and telemetry","h":"Feedback","x":"Feedback , at the top of the app, sends what you type to Michal, creator of Graspable.dev: what you are building, what got in the way, your words, and the app's version. Nothing from your projects, no prompts and no paths. If you tick Michal may reply to me by email , your account's address (or the one you type, when signed out) goes with it so he can answer.","u":"/docs/privacy-and-telemetry#feedback"},{"k":"docs","t":"Privacy and telemetry","h":"Your email after a first scene","x":"When your first scene finishes and you are not signed in, a small card from Michal offers free Credits for your email address. It is the same sign-in as in Settings, and it makes a free account. Close it and it does not come back. A new account gets a few emails from Michal over its first week, and every one has a link to stop them. When you sign in, the app also tells Graspable that it came from ","s":"When your first scene finishes and you are not signed in, a small card from Michal offers free Credits for your email address. It is the same sign-in as in Settings, and it makes a free account. Close it and it does not come back. A new account gets a few emails from Michal over its first week, and every one has a link to stop them. When you sign in, the app also tells Graspable that it came from the app, and the website tells it which tagged link or site first brought your browser there, so we know which places send people.","u":"/docs/privacy-and-telemetry#your-email-after-a-first-scene"},{"k":"docs","t":"Privacy and telemetry","h":"The pages written for agents","x":"When an assistant, search crawler or script requests a public reference on the website, Graspable counts its reported category and the public reference path. This includes the Markdown pages, agent discovery data, skill index and site search index. These counts contain no query parameters, IP address, account ID or prompt, and create no person profile. They count requests, not successful reads or ","s":"When an assistant, search crawler or script requests a public reference on the website, Graspable counts its reported category and the public reference path. This includes the Markdown pages, agent discovery data, skill index and site search index. These counts contain no query parameters, IP address, account ID or prompt, and create no person profile. They count requests, not successful reads or recommendations. A browser's ordinary page analytics is separate.","u":"/docs/privacy-and-telemetry#the-pages-written-for-agents"},{"k":"docs","t":"Privacy and telemetry","h":"Ratings","x":"Under every agent message there are copy, 👍 and 👎 buttons. A rating is something you choose to send. A rating sends the rating itself with the same kind of anonymized details as a summary: engine, model, instruction version, and what the agent had done so far. With 👎 you can pick reasons and type a comment, which is sent too. The agent's message and your request are sent only when you tick Incl","s":"Under every agent message there are copy, 👍 and 👎 buttons. A rating is something you choose to send. A rating sends the rating itself with the same kind of anonymized details as a summary: engine, model, instruction version, and what the agent had done so far. With 👎 you can pick reasons and type a comment, which is sent too. The agent's message and your request are sent only when you tick Include this response and my request . If you are signed out or offline, the rating is saved and sent later. The app tells you which.","u":"/docs/privacy-and-telemetry#ratings"},{"k":"docs","t":"Privacy and telemetry","h":"What is never sent","x":"Run summaries never contain: - your requests, - your code or file contents, - file names or paths, - the commands or other details of what the agent did, - your memory. API keys and sign-ins are never sent to Graspable.","u":"/docs/privacy-and-telemetry#what-is-never-sent"},{"k":"docs","t":"Privacy and telemetry","h":"Turn it off","x":"Open Settings → Help improve Graspable and untick Send run summaries, error reports and usage of the app . With it off, no summaries, no error reports and no usage events are sent from the app, and grasp and the MCP server stop counting their use too, and your runs are not placed in instruction tests. Ratings are separate: they are sent only when you click 👍 or 👎.","u":"/docs/privacy-and-telemetry#turn-it-off"},{"k":"docs","t":"Privacy and telemetry","h":"Where keys and sign-ins are stored","x":"What Where --- --- API keys for model providers Your system's keychain Your Graspable account sign-in Your system's keychain System One and cloud provider keys Your system's keychain ChatGPT sign-in for the Graspable agent A file in ~/.graspable/agent on your computer Claude Code sign-in Claude Code's own storage. Graspable never reads it Codex CLI sign-in The Codex CLI's own storage. Graspable ne","s":"What Where --- --- API keys for model providers Your system's keychain Your Graspable account sign-in Your system's keychain System One and cloud provider keys Your system's keychain ChatGPT sign-in for the Graspable agent A file in ~/.graspable/agent on your computer Claude Code sign-in Claude Code's own storage. Graspable never reads it Codex CLI sign-in The Codex CLI's own storage. Graspable never reads it Tokens of your cloud agents ~/.graspable/cloud.json , readable only by you A key is read from the keychain when a run starts and handed to the agent on your computer. For a cloud project, the key is sent with the run to your cloud agent over an encrypted connection and is not stored there.","u":"/docs/privacy-and-telemetry#where-keys-and-sign-ins-are-stored"},{"k":"docs","t":"Privacy and telemetry","h":"Other connections the app makes","x":"- Updates: the app checks for new versions when you ask in Settings → Updates . - Model lists: when it starts, the agent reads public lists of models (a model catalog, OpenRouter, Venice, Graspable for the models on Credits) and asks the public package registry for the newest version of Claude Code and the Codex CLI. These requests carry no key and nothing about you or your projects. - Agent spend","s":"- Updates: the app checks for new versions when you ask in Settings → Updates . - Model lists: when it starts, the agent reads public lists of models (a model catalog, OpenRouter, Venice, Graspable for the models on Credits) and asks the public package registry for the newest version of Claude Code and the Codex CLI. These requests carry no key and nothing about you or your projects. - Agent spending: when you switch it on, the agent asks a public catalogue for paid services and calls the services it buys from; payments go from the agent's wallet to the service. Graspable is not in that path and is told nothing about a purchase. The record of purchases is a file on your computer. See Agent spending. - Routstr, paid with bitcoin: to list nodes, the app asks public Nostr relays for their announcements and then asks each node about itself. When you use a node, your requests go to that node and on to the model provider it uses; what a node keeps is up to the people who run it. Nothing about your balance is sent to Graspable. - Credits paid with crypto: this is a purchase from Graspable, so Graspable keeps a record of it with your account: how many Credits, the coin and amount, the dollar value, the payment's reference on its network and, where the network shows it, the address that paid. To see whether your payment has arrived, Graspable asks public services about its own receiving address (block explorers, a price service, and for Lightning the service that issued the invoice). Your wallet is never asked for and never touched. - Venice, paid with crypto: when you use it, your requests go to Venice, signed by the agent's wallet. To show what is in that wallet, the app asks a public node of the Base network for the balance of its address. Nothing about the wallet is sent to Graspable. - Plans: Settings → Plan and Credits asks Graspable for the plan list and, when signed in, your plan. - Headset sharing: when you choose Share on Wi-Fi , the preview is reachable on your local network until you stop. See Preview and headset. - System One: when you are signed in, each thing it judges (a command, a saved note's title and text, a passage of an earlier run, a stretch of command output) is sent to Graspable and to our provider for that judgement, TypeSafe. Graspable stores how many were made, not what was judged. If you add your own model, it is sent there instead. Set the mode to Off to send nothing. See Approvals and safety. Missing AI account settings are shown in the app without sending a crash report. Other request failures remain reportable while Help improve Graspable is enabled.","u":"/docs/privacy-and-telemetry#other-connections-the-app-makes"},{"k":"docs","t":"Troubleshooting","x":"Fixes for common problems with installing, project setup, the preview, sign-in, checks, headsets and cloud agents. ","u":"/docs/troubleshooting"},{"k":"docs","t":"Troubleshooting","h":"The app will not open","x":"The macOS app is signed and checked by Apple. If macOS still refuses to open it, the download is probably incomplete: delete it, download it again from graspable.dev/download and compare its checksum with the one on the download card.","u":"/docs/troubleshooting#the-app-will-not-open"},{"k":"docs","t":"Troubleshooting","h":"\"agent unavailable\" on the Projects page","x":"The agent that runs beside the app did not start. Quit Graspable fully and open it again. If it keeps happening, install the latest version from graspable.dev/download.","u":"/docs/troubleshooting#quot-agent-unavailable-quot-on-the-projects-page"},{"k":"docs","t":"Troubleshooting","h":"\"install failed\" on a project","x":"Creating a project installs its packages. The project list shows the last line of the error. - No internet connection. Packages are downloaded during install. - You chose your own Node.js under Settings → Components → Advanced, and it is older than version 20. Update it, or switch back to Built into Graspable . A failed install does not lock the project. You can still send a request, for example t","s":"Creating a project installs its packages. The project list shows the last line of the error. - No internet connection. Packages are downloaded during install. - You chose your own Node.js under Settings → Components → Advanced, and it is older than version 20. Update it, or switch back to Built into Graspable . A failed install does not lock the project. You can still send a request, for example to ask the agent to fix package.json . The simplest fix is to create the project again.","u":"/docs/troubleshooting#quot-install-failed-quot-on-a-project"},{"k":"docs","t":"Troubleshooting","h":"The preview does not start or is empty","x":"- Preview and Split are disabled while packages are installing. Wait for \"installing deps\" to disappear from the header. - If starting fails, a red bar under the header shows the reason. After a failed install the preview cannot start. - If the preview stops responding, use the round-arrow button. It restarts the dev server if needed. - If the preview area stays empty in an older version, update. ","s":"- Preview and Split are disabled while packages are installing. Wait for \"installing deps\" to disappear from the header. - If starting fails, a red bar under the header shows the reason. After a failed install the preview cannot start. - If the preview stops responding, use the round-arrow button. It restarts the dev server if needed. - If the preview area stays empty in an older version, update. Version 0.9.15 fixed a blocked preview in the installed app.","u":"/docs/troubleshooting#the-preview-does-not-start-or-is-empty"},{"k":"docs","t":"Troubleshooting","h":"A sign-in did not finish","x":"- \"No model account yet\". The Graspable agent has no sign-in or key. Open Settings → AI for the agent . - The browser did not open. Choose Browser didn't open? Open the sign-in page or Open the sign-in page again . - ChatGPT. Choose Not working? Use a code instead and enter the code shown on the page that opened. - Claude. If the page shows a code, paste it into the field in Settings. - Graspable ","s":"- \"No model account yet\". The Graspable agent has no sign-in or key. Open Settings → AI for the agent . - The browser did not open. Choose Browser didn't open? Open the sign-in page or Open the sign-in page again . - ChatGPT. Choose Not working? Use a code instead and enter the code shown on the page that opened. - Claude. If the page shows a code, paste it into the field in Settings. - Graspable account. Open the email link on the same computer. If the email shows a code, type it into the field. - \"ChatGPT sign-in has ended\" or \"Claude is not signed in, or the session expired\". Sign in again in Settings. Choose Cancel to start over. You never need a terminal to sign in.","u":"/docs/troubleshooting#a-sign-in-did-not-finish"},{"k":"docs","t":"Troubleshooting","h":"Old changes or approvals reappear after a connection interruption","x":"The next desktop update fixes repeated file updates and answered approval cards after a run reconnects. It also keeps late responses from appearing in a different conversation. Updated agents resume after the last received update instead of sending the whole conversation again; cloud projects need the updated Cloud Agent as well. If you see this in an older version, reopen the project to load its ","s":"The next desktop update fixes repeated file updates and answered approval cards after a run reconnects. It also keeps late responses from appearing in a different conversation. Updated agents resume after the last received update instead of sending the whole conversation again; cloud projects need the updated Cloud Agent as well. If you see this in an older version, reopen the project to load its saved conversation and files.","u":"/docs/troubleshooting#old-changes-or-approvals-reappear-after-a-connection-interruption"},{"k":"docs","t":"Troubleshooting","h":"The conversation stops updating after opening history","x":"The next desktop update keeps the current conversation live if another conversation cannot load. Slow history requests also stop replacing a newer selection. If you switch conversations while a new request is still preparing, that preparation will not start an agent run. Requests already accepted by the agent continue until they finish or you cancel them.","u":"/docs/troubleshooting#the-conversation-stops-updating-after-opening-history"},{"k":"docs","t":"Troubleshooting","h":"Typing feels slow in a long conversation","x":"The next desktop update avoids redrawing unchanged responses as you type. Your full conversation remains available to read. Switching conversations also closes feedback forms from the previous conversation.","u":"/docs/troubleshooting#typing-feels-slow-in-a-long-conversation"},{"k":"docs","t":"Troubleshooting","h":"A conversation cannot reconnect","x":"In the next desktop update, a connection error offers Retry above the conversation. Retry loads that same run and checks its current state; it does not submit your prompt again. A failed history load leaves the visible conversation available. If a run was interrupted, retry can continue it after reconnecting. Your unsent text stays in the prompt box. If the conversation list itself could not load,","s":"In the next desktop update, a connection error offers Retry above the conversation. Retry loads that same run and checks its current state; it does not submit your prompt again. A failed history load leaves the visible conversation available. If a run was interrupted, retry can continue it after reconnecting. Your unsent text stays in the prompt box. If the conversation list itself could not load, Retry checks that list again and opens the latest conversation. A successful empty history is not treated as an error.","u":"/docs/troubleshooting#a-conversation-cannot-reconnect"},{"k":"docs","t":"Troubleshooting","h":"A request could not be sent","x":"In the next desktop update, a failed or abandoned submission keeps your prompt and selected code so you can try again. If Graspable says the request started but its updates could not open, use Retry to reconnect instead of sending the same request again. Text you type while a previous request is being sent stays in the prompt box. Code you select during that wait stays selected for your next reque","s":"In the next desktop update, a failed or abandoned submission keeps your prompt and selected code so you can try again. If Graspable says the request started but its updates could not open, use Retry to reconnect instead of sending the same request again. Text you type while a previous request is being sent stays in the prompt box. Code you select during that wait stays selected for your next request; only the selection sent with the accepted request is cleared.","u":"/docs/troubleshooting#a-request-could-not-be-sent"},{"k":"docs","t":"Troubleshooting","h":"A check keeps failing","x":"After three repair turns the run stops and names the check. - build — read the error in the transcript. Tell the agent what you see, or undo the run. - smoke — look at the screenshot under the check. It shows what the browser saw. - xr … differ from the recording — the scene changed. If you meant it, choose Accept new baseline . See XR tests. - \"skipped: there is no browser to load the scene in ye","s":"After three repair turns the run stops and names the check. - build — read the error in the transcript. Tell the agent what you see, or undo the run. - smoke — look at the screenshot under the check. It shows what the browser saw. - xr … differ from the recording — the scene changed. If you meant it, choose Accept new baseline . See XR tests. - \"skipped: there is no browser to load the scene in yet\" — open Settings → Components . It shows the browser download, or Try again if it failed. Interrupted or oversized downloads are discarded; a stalled download stops with an error instead of waiting indefinitely. The build check still runs.","u":"/docs/troubleshooting#a-check-keeps-failing"},{"k":"docs","t":"Troubleshooting","h":"\"The recording is empty\"","x":"An XR recording has no movement in it. This happens when the app crashed in XR while it was being recorded. Record it again, or delete the file from tests/xr/ .","u":"/docs/troubleshooting#quot-the-recording-is-empty-quot"},{"k":"docs","t":"Troubleshooting","h":"The run stopped by itself","x":"- \"budget\" — the run crossed a cap in Settings → Run budget . - A ⚠ line about a loop or repeated failures — the agent was stuck. Rephrase the request or split it into smaller steps. - \"continuing after the restart\" — the app restarted while the Graspable agent was working (an update, a crash, a forced quit). Nothing is lost: the run goes on from its last step as soon as you open the project. It a","s":"- \"budget\" — the run crossed a cap in Settings → Run budget . - A ⚠ line about a loop or repeated failures — the agent was stuck. Rephrase the request or split it into smaller steps. - \"continuing after the restart\" — the app restarted while the Graspable agent was working (an update, a crash, a forced quit). Nothing is lost: the run goes on from its last step as soon as you open the project. It asks again for anything it was waiting on. A run of Claude Code or the Codex CLI cannot be continued this way and is closed as failed.","u":"/docs/troubleshooting#the-run-stopped-by-itself"},{"k":"docs","t":"Troubleshooting","h":"Voice","x":"- \"Graspable may not use the microphone\" : allow Graspable to use the microphone in your system's privacy settings, then try again. - \"No microphone was found\" : connect a microphone or turn one on, then press the button again. - \"The microphone is in use by another app\" : close the app that holds it, or end the call, then try again. - \"Nothing was heard\" : speak a little closer to the microphone.","s":"- \"Graspable may not use the microphone\" : allow Graspable to use the microphone in your system's privacy settings, then try again. - \"No microphone was found\" : connect a microphone or turn one on, then press the button again. - \"The microphone is in use by another app\" : close the app that holds it, or end the call, then try again. - \"Nothing was heard\" : speak a little closer to the microphone. A press shorter than half a second is ignored. - \"Sign in to your Graspable account to use voice\" : sign in under Settings → Graspable account , or add your own ElevenLabs key under Settings → Voice . - \"ElevenLabs did not accept the key\" : check the key under Settings → Voice . - \"Today's free voice is used up\" : voice beyond the free minutes uses Credits. Add Credits, or use your own ElevenLabs key. See Credits. - \"The recording is too long\" : one recording holds a few minutes of speech. Say the request in shorter parts. - The words come out in the wrong language : choose the language you speak under Settings → Voice . - The button in the headset says \"Reconnect voice from Graspable\" : allow the microphone in the headset's browser. Then choose Talk in the headset again in the app and scan the new code. See Voice.","u":"/docs/troubleshooting#voice"},{"k":"docs","t":"Troubleshooting","h":"Headset","x":"- \"No network connection\" or the link does not load — put the computer and the headset on the same Wi-Fi, or use USB. If the panel offers several addresses, try another. - The headset warns that the connection is not private — expected the first time. Choose Advanced , then Proceed . - USB shows no headset — install Android platform-tools ( adb ), turn on developer mode in the headset, plug it in,","s":"- \"No network connection\" or the link does not load — put the computer and the headset on the same Wi-Fi, or use USB. If the panel offers several addresses, try another. - The headset warns that the connection is not private — expected the first time. Choose Advanced , then Proceed . - USB shows no headset — install Android platform-tools ( adb ), turn on developer mode in the headset, plug it in, then allow USB debugging inside the headset. - Cloud project — headset preview is not available. Move the project back to this computer.","u":"/docs/troubleshooting#headset"},{"k":"docs","t":"Troubleshooting","h":"Cloud agents","x":"Message What to do --- --- \"The cloud agent runs Graspable …, which this app cannot work with\" The Cloud Agent is from a much older version. Click Cloud in the project header and choose Update the cloud agent \"You have no Credits left\" Cloud Agents run on Credits. Buy a Credit pack, turn on overage, or choose a plan. See Credits \"This template comes with the Builder plan and higher\" (or \"the Pro p","s":"Message What to do --- --- \"The cloud agent runs Graspable …, which this app cannot work with\" The Cloud Agent is from a much older version. Click Cloud in the project header and choose Update the cloud agent \"You have no Credits left\" Cloud Agents run on Credits. Buy a Credit pack, turn on overage, or choose a plan. See Credits \"This template comes with the Builder plan and higher\" (or \"the Pro plan and higher\") Premium templates need a paid plan; the native XR ones need Pro. Sign in under Settings → Graspable account and go online once, so the app can check your plan \"You have used this month's … Credits\" Wait for them to renew, upgrade, or move the project back to your computer. See Credits \"Cloud Agents are not ready for Graspable … yet\" Update the app, then try again \"The cloud agent is Graspable X and this app is Y. They must match\" Update the app, or the image or server, so both are the same version \"The cloud agent refused the token\" Your own server: check the Agent token in Settings → Cloud Agents \"The cloud agent did not start in time\" Try again. For your own server, check that it is running and reachable \"The cloud agent cannot be reached\" when moving back Choose Wake now , then try again The header shows \"not found\" The sandbox no longer exists. Choose Restore the copy from before the move A run is refused in the cloud Cloud runs need the Graspable agent with an API key \"… its cloud sandbox could not be deleted\" The project is safely back. Delete the sandbox in your provider's dashboard","u":"/docs/troubleshooting#cloud-agents"},{"k":"docs","t":"Troubleshooting","h":"Confirming an action","x":"Deleting a project or capture, changing seats, and discarding animation or USD edits wait for your answer in a confirmation dialog. Choose Cancel to keep things as they are.","u":"/docs/troubleshooting#confirming-an-action"},{"k":"docs","t":"Troubleshooting","h":"Still stuck","x":"Tell us on Discord. Include the app version from Settings → Updates and the text of the error.","u":"/docs/troubleshooting#still-stuck"},{"k":"docs","t":"Frequently asked questions","x":"Short answers about cost, privacy, subscriptions, headsets and what you need to use Graspable. ","u":"/docs/faq"},{"k":"docs","t":"Frequently asked questions","h":"What is Graspable?","x":"Graspable is an AI app builder for XR: virtual reality, augmented reality and mixed reality. Describe an experience and its desktop agent creates an editable project on your computer. Web projects use React Three Fiber and WebXR. Native templates use Unity, Unreal Engine or Swift for visionOS and require the corresponding development tools. See Templates, Native XR and the tool comparisons.","u":"/docs/faq#what-is-graspable"},{"k":"docs","t":"Frequently asked questions","h":"Is Graspable free?","x":"Yes, for building on your own computer with your own AI account. You pay your AI provider, not Graspable. Paid plans add Cloud Agents, premium templates and the Credits to run them. See Plans and billing.","u":"/docs/faq#is-graspable-free"},{"k":"docs","t":"Frequently asked questions","h":"How much does Graspable cost?","x":"Explorer is free. On monthly billing, Builder costs $20, Pro $40, and Studio $99 per seat with a minimum of three seats ($297 a month). Annual billing charges the year upfront; the pricing page shows the totals, included Credits and publishing limits. Your AI subscription or API usage is a separate cost when you bring your own account. Premium web templates need Builder or above; native XR templat","s":"Explorer is free. On monthly billing, Builder costs $20, Pro $40, and Studio $99 per seat with a minimum of three seats ($297 a month). Annual billing charges the year upfront; the pricing page shows the totals, included Credits and publishing limits. Your AI subscription or API usage is a separate cost when you bring your own account. Premium web templates need Builder or above; native XR templates and the OpenUSD asset library need Pro or Studio. Hosted work uses Credits. See Plans and billing and Credits before choosing a plan.","u":"/docs/faq#how-much-does-graspable-cost"},{"k":"docs","t":"Frequently asked questions","h":"Do I need to know how to code?","x":"No. You describe what you want and the agent writes the code. The code is a normal project you can read and edit, so knowing some code helps when you want fine control.","u":"/docs/faq#do-i-need-to-know-how-to-code"},{"k":"docs","t":"Frequently asked questions","h":"Can I talk to the agent instead of typing?","x":"Yes. The microphone beside the prompt turns what you say into the text of a request. You can also start a conversation: the agent listens, takes what you said when you stop speaking, and reads its answers aloud. In VR or AR, a Talk button does the same from inside the scene. Speech is handled by ElevenLabs, through your Graspable account on Credits with some free minutes every day, or with your ow","s":"Yes. The microphone beside the prompt turns what you say into the text of a request. You can also start a conversation: the agent listens, takes what you said when you stop speaking, and reads its answers aloud. In VR or AR, a Talk button does the same from inside the scene. Speech is handled by ElevenLabs, through your Graspable account on Credits with some free minutes every day, or with your own ElevenLabs key. See Voice.","u":"/docs/faq#can-i-talk-to-the-agent-instead-of-typing"},{"k":"docs","t":"Frequently asked questions","h":"Does my code leave my computer?","x":"The project lives and runs on your computer. Your requests and the code the agent reads go to the model provider you chose, so the model can answer. Graspable receives only counts-only run summaries and ratings you send, never your code. See Privacy and telemetry.","u":"/docs/faq#does-my-code-leave-my-computer"},{"k":"docs","t":"Frequently asked questions","h":"Can I use my ChatGPT subscription?","x":"Yes. Select Graspable agent in Settings → AI for the agent and choose Sign in with ChatGPT . OpenAI tolerates this use but does not officially support it, so it may stop working. An API key is the fallback.","u":"/docs/faq#can-i-use-my-chatgpt-subscription"},{"k":"docs","t":"Frequently asked questions","h":"Can I use my Claude subscription?","x":"Yes. Choose Graspable agent → Claude subscription in Settings to keep Graspable’s tools and workflow, or choose Claude Code to run its own agent. The Graspable option is experimental and stays on this computer. Usage follows your Claude subscription and Agent SDK allowance; extra usage depends on your account settings. See Engines and sign-in.","u":"/docs/faq#can-i-use-my-claude-subscription"},{"k":"docs","t":"Frequently asked questions","h":"Can I use a local model?","x":"Yes. Add an account of the kind Custom / local (OpenAI-compatible) and point it at your endpoint, for example Ollama or LM Studio. See Engines and sign-in.","u":"/docs/faq#can-i-use-a-local-model"},{"k":"docs","t":"Frequently asked questions","h":"Which headsets work?","x":"The preview opens in a headset's own browser, so the headset needs a browser with WebXR. Over Wi-Fi you scan a QR code. Over USB, the app supports Quest, Pico and Android XR headsets. You can also build without a headset, using the emulated one in the preview. See Preview and headset.","u":"/docs/faq#which-headsets-work"},{"k":"docs","t":"Frequently asked questions","h":"What does the agent build with?","x":"Web projects use Vite, React, React Three Fiber and @react-three/xr . The world templates add Gaussian splats, shown with the Spark renderer, and places generated by World Labs. Native templates use Unity, Unreal Engine or Swift for visionOS and need the corresponding engine tools. See Templates and Native XR.","u":"/docs/faq#what-does-the-agent-build-with"},{"k":"docs","t":"Frequently asked questions","h":"Can the agent break my computer?","x":"File tools work only inside the project folder. Running a command asks you first, and you see the command before you allow it. Every file change is recorded, and Undo run puts the files back. See Approvals and safety.","u":"/docs/faq#can-the-agent-break-my-computer"},{"k":"docs","t":"Frequently asked questions","h":"What happens when the agent gets it wrong?","x":"Graspable checks web projects in a browser and native projects with their engine tools. If the checks fail, the agent gets up to three repair attempts. If you do not like the result, undo the run or tell the agent what to change. Device testing is still needed for interactions and performance on your target hardware.","u":"/docs/faq#what-happens-when-the-agent-gets-it-wrong"},{"k":"docs","t":"Frequently asked questions","h":"What if the app restarts during a run?","x":"With the Graspable agent, nothing is lost. Every step of the conversation, every model answer and every tool call is written down as it happens, so when the app comes back and you open the project, the run continues from its last step. If it was waiting for your yes, it asks again. The checks and the repair turns go on from where they were. Claude Code and the Codex CLI keep their own sessions; a ","s":"With the Graspable agent, nothing is lost. Every step of the conversation, every model answer and every tool call is written down as it happens, so when the app comes back and you open the project, the run continues from its last step. If it was waiting for your yes, it asks again. The checks and the repair turns go on from where they were. Claude Code and the Codex CLI keep their own sessions; a run of theirs that the app interrupted is closed, and your next request starts with a short account of what it did.","u":"/docs/faq#what-if-the-app-restarts-during-a-run"},{"k":"docs","t":"Frequently asked questions","h":"Do I need a Graspable account?","x":"No. An account is needed only for plans, for what runs on Credits (such as voice without your own ElevenLabs key), and for sending ratings and anonymized run summaries.","u":"/docs/faq#do-i-need-a-graspable-account"},{"k":"docs","t":"Frequently asked questions","h":"Can a project keep running when my computer is off?","x":"Yes, if you move it to the cloud. Cloud Agents use Credits: a new account gets 6 once to try one, and paid plans include Credits every month. Cloud projects need the Graspable agent with an API key. See Cloud Agents.","u":"/docs/faq#can-a-project-keep-running-when-my-computer-is-off"},{"k":"docs","t":"Frequently asked questions","h":"Where are my projects?","x":"In ~/Graspable/projects , one normal folder per project, so you can open them in any editor or with git. To keep them somewhere else, choose a folder under Settings → Projects folder ; Graspable moves the existing projects there and restarts. See Install Graspable.","u":"/docs/faq#where-are-my-projects"},{"k":"changelog","t":"Talk with the agent in the headset","h":"Version 0.21.3","x":"Press Talk in VR or AR and speak: the agent hears you and answers in the headset, in the preview on your computer and on a real headset over Wi-Fi. Voice in XR: a Talk button low in front of you starts a conversation with the agent. In the preview on your computer it uses this computer's microphone. Headset: Talk in the headset lets a real headset listen with its own microphone and play the agent'","u":"/changelog#v0.21.3"},{"k":"changelog","t":"Talk with the agent","h":"Version 0.21.2","x":"Start a conversation and the agent listens, takes your request when you stop speaking, and reads its answers aloud. Speak to interrupt it. Voice: the button with the sound waves starts a conversation. What you say is sent when you stop speaking, as a new request or as a note to the agent while it works, and its answers are read aloud in a few short sentences. Speaking while it reads stops the read","u":"/changelog#v0.21.2"},{"k":"changelog","t":"Speak your requests","h":"Version 0.21.1","x":"Press the microphone beside the prompt and say what you want. Speech is turned into text by ElevenLabs, on Credits or with your own key. Voice: the microphone beside the prompt turns what you say into text in the prompt, where you can change it before sending. Escape drops the recording. Settings, Voice: choose Graspable on Credits (about ten minutes free every day) or your own ElevenLabs key, the","u":"/changelog#v0.21.1"},{"k":"changelog","t":"Design","h":"Version 0.21.0","x":"Design, the editor for the panels in your 3D scene, is complete: edit by hand or with the agent, move with springs, check for each device, and write one panel for Apple Vision Pro, Unity headsets and Meta Ray-Ban Display. Design brings together what came in 0.20.2 to 0.20.10: a flat view of each panel, live handles for size, gaps and padding, springs, a theme, device checks, the Animate timeline, ","u":"/changelog#v0.21.0"},{"k":"changelog","t":"Animate panels, and Design in any web project","h":"Version 0.20.10","x":"Panels can be animated on the Animate timeline, the preview follows a drag in Design, the UI kit has new parts, and any web project can add the kit. Design: Add the UI kit puts the panels, buttons, sliders and the rest into a project made from another web template, in a folder of its own, and asks the agent for a first panel. Tabletop AR, React XR, product viewers and games can then be designed li","u":"/changelog#v0.20.10"},{"k":"changelog","t":"Theme colours and spacing","h":"Version 0.20.9","x":"Design edits the theme every part uses, and lets a part follow a theme colour. Design: with nothing selected, the inspector shows the theme: colours, spacing, how controls move and how panels come in. A change there reaches every part that uses it. Design: give a part a theme colour from the list beside its colour, so it changes with the theme, or a colour of its own.","u":"/changelog#v0.20.9"},{"k":"changelog","t":"One panel, several platforms","h":"Version 0.20.8","x":"Design writes a panel as SwiftUI for Apple Vision Pro, as Unity UI Toolkit for headsets built with Unity, and as a screen for Meta Ray-Ban Display. Design: Use on another platform. A selected panel is written as a SwiftUI view for Apple Vision Pro, as a UXML layout and style sheet for Unity UI Toolkit, or as a screen of a Display glasses app, ready to copy. Sliders and toggles bound to state becom","u":"/changelog#v0.20.8"},{"k":"changelog","t":"Checks for headsets and phones","h":"Version 0.20.7","x":"Design tells you whether an interface can be read and used where it stands, in a headset or on a phone. Design: Checks for the interface you are looking at. For a headset, whether a finger can reach it, whether its text is large enough for its distance, whether its controls are big enough for a fingertip or for pointing, and whether it sits comfortably in view. For phone AR, text and controls in p","u":"/changelog#v0.20.7"},{"k":"changelog","t":"Faster design edits","h":"Version 0.20.6","x":"Drag gaps, padding and parts into place, and design with the keyboard. Design: select a row, column or panel to see the space between its parts. Drag it to change the gap, or drag a panel's edges to change its padding, with the value shown as you drag. Design: drag parts in the list to reorder them, or drop one onto a row, column or panel to put it inside. Design: keys for undo, redo, delete, dupl","u":"/changelog#v0.20.6"},{"k":"changelog","t":"Springs for buttons and panels","h":"Version 0.20.5","x":"Buttons, toggles and sliders move with springs, panels come in with a short movement, and Design lets you choose and preview both. Design: a Motion section for buttons, toggles, sliders and panels. Choose a spring (Quick, Gentle, Bouncy, Slow or None) and see its curve, choose how a panel comes in (Rise, Pop or Tilt), and press Play to see it in the preview. In new Hands & Spatial UI projects, con","u":"/changelog#v0.20.5"},{"k":"changelog","t":"Parts by name","h":"Version 0.20.4","x":"Design names each part by its words or colour, and shows the size while you drag a handle. Design: the parts list names sliders, toggles and parts made from the same line of code by their words or colour, such as \"Speed\" or \"colour #ff7a59\", instead of only their kind. Design: while you drag a handle, the new width and height show under the part before you let go.","u":"/changelog#v0.20.4"},{"k":"changelog","t":"Resize parts by hand","h":"Version 0.20.3","x":"Design: drag the corners of a selected part to size it, and edit values without losing your place. Design: drag the handles on a selected button, slider, toggle, label or panel in the flat view to change its width and height. Design: changing a value keeps the inspector where you scrolled it, and keeps the field you are typing in, so the arrow keys can step a number again and again. Design: select","u":"/changelog#v0.20.3"},{"k":"changelog","t":"Design panels and controls","h":"Version 0.20.2","x":"Design: an editor for the panels, buttons and sliders inside a 3D scene, shown flat and straight on, that writes every change into the code. Design: an editor for the panels inside a 3D scene. Each panel is shown flat and straight on, with its parts listed beside it. Click a button, slider, toggle or label and change its text, size, colours and layout in an inspector; add, move, copy and remove pa","u":"/changelog#v0.20.2"},{"k":"changelog","t":"WebGPU in the headset","h":"Version 0.20.1","x":"WebGPU Splats scenes draw on WebGPU in a headset whose browser offers it, and checks on a Mac no longer set off a macOS warning. In a headset browser that offers WebGPU for immersive sessions, the WebGPU Splats template now draws both eyes on WebGPU. Other headsets keep the WebGL2 path. On a Mac, checks now use their own copy of Chrome for Testing, downloaded once when Graspable starts. macOS no l","u":"/changelog#v0.20.1"},{"k":"changelog","t":"Splats made by code, done properly","h":"Version 0.20.0","x":"The WebGPU Splats template now comes with a guide the agent reads before writing splat code, and its renderer setup is tested piece by piece. Projects from the WebGPU Splats template include a guide for the agent on splats made by code: how to write a generator, how to trace an expensive shape into a table on the graphics card a slice at a time, how many splats each kind of device draws, and the m","u":"/changelog#v0.20.0"},{"k":"changelog","t":"WebGPU Splats","h":"Version 0.19.9","x":"A new template draws millions of Gaussian splats on WebGPU, from capture files and from code that runs on the graphics card. The WebGPU Splats template draws Gaussian splats with Radix, the WebGPU splat renderer from the makers of Spark. Every splat in the scene is sorted and drawn together on the graphics card, so a recent computer shows one to two million smoothly. It starts with a galaxy of 600","u":"/changelog#v0.19.9"},{"k":"changelog","t":"Gestures, the real room, and AI inside your apps","h":"Version 0.19.8","x":"Web projects recognise hand and head gestures, physics meets the real room, and published apps can ask an AI model with no key in the app. The React XR and XR Game templates recognise hand gestures (pinch, point, fist, thumbs-up, open palm, spread) and head gestures (nod and shake). In a new React XR project a thumbs-up adds a ball. In AR, the React XR template turns the surfaces and room shapes t","u":"/changelog#v0.19.8"},{"k":"changelog","t":"Helpers work at the same time","h":"Version 0.19.7","x":"Helpers the Graspable agent starts in one step now work side by side. When the Graspable agent starts several helpers in one step, they now work at the same time instead of one after the other. Other tools still run one at a time.","u":"/changelog#v0.19.7"},{"k":"changelog","t":"USD scenes open directly","h":"Version 0.19.6","x":"USD scenes open directly in the asset viewer without conversion, and payment details are easier to copy. USD scenes open directly in the asset viewer, read by Hugging Face's OpenUSD web viewer, with no conversion and no extra download. A single SimReady asset opens in about half a second. Scenes with animation offer Play its animation, which opens the converted preview. Installer checksums confirm","u":"/changelog#v0.19.6"},{"k":"changelog","t":"Open a scene, see the model","h":"Version 0.19.5","x":"USD previews open automatically, with simpler editing controls and more private error reports. USD scenes open with an automatic 3D preview in the asset sidebar and a larger scene viewer. Editing and export tools stay out of the way until needed; previews reuse unchanged scenes without adding project files. Fit model brings the current animated pose into view. Restore pose keeps the camera unchang","u":"/changelog#v0.19.5"},{"k":"changelog","t":"Reliable browser setup","h":"Version 0.19.4","x":"Recover from interrupted downloads and run browser checks on a fresh Linux installation. Browser setup recovers from interrupted downloads and file-write failures without stopping the agent. Stalled or oversized transfers stop with a retryable error. First-time browser setup on Linux no longer needs a separate archive tool. Linux installers supply the extra libraries used by browser checks.","u":"/changelog#v0.19.4"},{"k":"changelog","t":"Keep your work, edit more smoothly","h":"Version 0.19.3","x":"Protect unsaved edits, recover interrupted conversations, and spend less time waiting on long chats, large timelines and asset folders. Closing or quitting asks before discarding unsaved code or animation, including Command-Q on Mac. Cancelling keeps the agent running. Downloaded updates wait for saved edits before installing. Typing stays responsive in long conversations, and large animation time","u":"/changelog#v0.19.3"},{"k":"changelog","t":"More reliable previews and checks","h":"Version 0.19.2","x":"Avoid duplicate preview servers and unnecessary restarts, use less memory when checking large assets, and keep file and download checks strict. Browser checks reject malformed requests and linked files outside the build. Large assets stream with less temporary memory, and interrupted transfers close cleanly. Opening a preview or running checks at the same time no longer starts duplicate background","u":"/changelog#v0.19.2"},{"k":"changelog","t":"Smoother editing and lighter previews","h":"Version 0.19.1","x":"Keep your edits when switching files, work more smoothly with large timelines, and browse Scenario models and splats in the Pro asset library. Saved code updates appear without an extra typing delay. Numeric dragging can resolve numbers beyond the initially parsed part of a large file. Large animation takes respond faster when moving between keys. Unchanged keyframe controls stay in place as the p","u":"/changelog#v0.19.1"},{"k":"changelog","t":"OpenUSD assets, ready to work with","h":"Version 0.19.0","x":"Search NVIDIA SimReady assets, keep their editable source, and build with them in Pro. A curated skills library gives your agent more ways to help. Pro and Studio include a searchable NVIDIA SimReady OpenUSD library from Hugging Face. Inspect package previews, sizes and licenses, then import source assets with attribution. Agents can search and import the same OpenUSD library. Downloads are checke","u":"/changelog#v0.19.0"},{"k":"changelog","t":"Confirmation dialogs and error reports","h":"Version 0.18.1","x":"Confirmation dialogs wait for your answer, and expired approval cards close cleanly. The Meta glasses guide helps you choose between a phone companion, a Display web app and an immersive VR project. The WebXR testing guide separates desktop emulation from real-headset checks and makes emulator setup opt-in. The Replit comparison now covers its documented WebXR workflow and current testing tools. A","u":"/changelog#v0.18.1"},{"k":"changelog","t":"Tabletop AR and Claude in the Graspable agent","h":"Version 0.18.0","x":"Start with an animated AR rescue film and puzzle, or use your Claude subscription with the Graspable agent. Agents in browsers offering experimental WebMCP can search the site and read public guides, templates, comparisons and pricing. Site search includes pricing and tool comparisons. The mobile download form and its email now include Windows when explaining where to open your link. A Graspable s","u":"/changelog#v0.18.0"},{"k":"changelog","t":"Transform keys and easier easing edits","h":"Version 0.17.7","x":"Edit a complete object pose together, adjust easing beside each channel, and use updated confidential image generation. Edit Position, Rotation and Scale together in Animate. Key one channel or save the complete pose with Key all transforms. Easing controls sit beside animation channels and camera moves. Open them when needed, including in the experimental XR editing panel. Packaged apps include t","u":"/changelog#v0.17.7"},{"k":"changelog","t":"Character performances and confidential images","h":"Version 0.17.6","x":"Edit facial expressions, speech and bone rotations, and generate project images through confidential Routstr. Key named blendshapes in Animate, with curves, retiming, save and undo. Matching controls on the face and teeth move together. Import timed speech cues, map them to a character’s mouth controls and edit the resulting keyframes. Existing mouth animation is protected unless you explicitly re","u":"/changelog#v0.17.6"},{"k":"changelog","t":"Audition characters and publish more reliably","h":"Version 0.17.5","x":"Check imported clips and facial controls before using a model. Publishing retries interrupted file uploads. Play, pause and scrub imported GLB and glTF animation clips in Assets. Test existing morph controls, reset the pose and frame the model. Preview changes leave your files untouched. Asset previews have a Close button and scroll within the sidebar to keep the asset list reachable. Model previe","u":"/changelog#v0.17.5"},{"k":"changelog","t":"Faster animation previews and safer publishing","h":"Version 0.17.4","x":"Compare templates, edit animation with less repeated work, and keep local configuration out of published source. Compare template screenshots, requirements and example requests in the new in-app picker. Template docs link to each template page. Animation playback reuses sampled poses, keyframe lists and curve paths instead of rebuilding them on every display refresh. OpenTimelineIO import planning","u":"/changelog#v0.17.4"},{"k":"changelog","t":"Animate scenes and share immersive films","h":"Version 0.17.2","x":"Edit animation, camera shots and sound, exchange supported USD scenes, and share films on the web or in a headset. An early release of the filmmaking workspace, with more work ahead for 0.18.0. Render MP4, mono 360° and side-by-side stereo films with sound. Supported Macs can also encode Apple spatial and projected 360° video. Publish a saved film on graspable.cloud with a screen and headset playe","u":"/changelog#v0.17.2"},{"k":"changelog","t":"Inspect motion at an exact frame","h":"Version 0.17.1","x":"An early tool for agents to check an animation document and inspect a pose at a chosen frame. This is the first step toward shared animation authoring; it does not yet play the animation in your preview. Agents can inspect position, scale, rotation, opacity and visibility tracks in a project motion file. Invalid keyframes and conflicting tracks are reported before a document is sampled. Seeking in","u":"/changelog#v0.17.1"},{"k":"changelog","t":"Agent payments from your own wallet","h":"Version 0.17.0","x":"This release brings together agent spending, crypto-paid AI and the latest fixes. Let your agent pay for outside services from its own wallet, within limits you set. These features remain experimental. Agent spending brings together limits for each purchase, request and day, approved and blocked services, and a separate question when a service changes its payment address. Teams can choose which se","u":"/changelog#v0.17.0"},{"k":"changelog","t":"Published apps as cards, and the welcome questions in a dialog","h":"Version 0.16.14","x":"Your published apps now show in the app as they do on the site, with pictures and counts. The welcome questions moved into a dialog, take several answers each, and are kept with your account. Settings, Published apps: each app is a card with its picture, what it is, its views and likes and its address, with totals above. What changes an app is under Manage. The welcome questions come up in a dialo","u":"/changelog#v0.16.14"},{"k":"changelog","t":"New models the day they come out, and a model menu you can search","h":"Version 0.16.13","x":"Model lists now renew themselves each time Graspable starts, so a new model needs no new version of the app. The Graspable agent on Credits offers every OpenRouter model that can use tools. Every model menu can be searched and filtered, and a project can run on its own model, chosen above the prompt. Meta's Muse models, Muse Wonder included, with an OpenRouter key or on Credits. Tried with an Open","u":"/changelog#v0.16.13"},{"k":"changelog","t":"Routstr confidential mode follows the protocol's update","h":"Version 0.16.12","x":"The author of the confidential mode changed the protocol in a way older clients cannot speak, and hardened its checks. This release moves Graspable to the new version. Update to keep using confidential mode on an updated node. It stays experimental. Confidential mode works again with a Routstr node that runs the protocol's current version. A request from an older Graspable is refused there and is ","u":"/changelog#v0.16.12"},{"k":"changelog","t":"Routstr in confidential mode","h":"Version 0.16.11","x":"On a Routstr node that offers it, the agent can now talk to the model provider through a connection the node relays and cannot read. It uses Juraj Bednar's protocol and his open-source client. Experimental, slower, and tried on one node with one provider. Confidential mode, a switch under a Routstr balance: the node cannot read or change your requests and the answers, and each request shows what w","u":"/changelog#v0.16.11"},{"k":"changelog","t":"A team's rule for agent spending, and a list of services to look through","h":"Version 0.16.10","x":"A team's owner and admins can now decide which outside services the team's agents may pay. And the public catalogue of paid services can be searched in Settings before you approve anything. Agent spending stays experimental. Your team's rule, under Settings, Agent spending: the owner and admins choose whether members decide for themselves or the team's agents pay only services approved for the tea","u":"/changelog#v0.16.10"},{"k":"changelog","t":"Agent spending: who it may pay, apart from how much","h":"Version 0.16.9","x":"You can now let the agent pay only services you approved. A new service asks you once, whatever it costs, and that approval is kept apart from approving a purchase. Agent spending stays experimental. Only services I approve, under Settings, Agent spending: a service that is not on your list asks you first, even inside the budget. Your yes approves the service, not a purchase. When nobody is there ","u":"/changelog#v0.16.9"},{"k":"changelog","t":"Long runs on Credits no longer stop after an hour","h":"Version 0.16.8","x":"A request the Graspable agent works on for more than an hour used to lose its AI account when your sign-in token ran out. Each run now has a key of its own. The Graspable agent on Credits keeps working through a long request. A run gets its own key for the model, good for twelve hours, where it used to depend on a sign-in token that lasts one.","u":"/changelog#v0.16.8"},{"k":"changelog","t":"Voice tools for Display glasses web apps","h":"Version 0.16.7","x":"A web app for Meta Ray-Ban Display can now offer its actions to Meta AI, works without a connection once loaded, and is hosted by publishing it. The voice part is a developer preview on Meta's side and has not been tried on the glasses. Voice tools in the Display glasses web app template: each screen offers its actions to Meta AI (\"Hey Meta, set the timer to three minutes\"), through Meta's WebMCP ","u":"/changelog#v0.16.7"},{"k":"changelog","t":"Android phones, and menus that close","h":"Version 0.16.5","x":"Run on my phone now takes an Android phone over its cable. Menus in the project header close when you click elsewhere. The glasses emulator is no longer inside your project. Run on my phone builds for an Android phone too, with Android Studio on the computer. The connection runs over the cable, so the phone need not share the Wi-Fi. A new Meta glasses project no longer carries the emulator's code:","u":"/changelog#v0.16.5"},{"k":"changelog","t":"Run on my phone","h":"Version 0.16.4","x":"A glasses app goes from the preview to your iPhone with one button, and then follows the agent's edits live. Tried on an iPhone 15 Pro with Ray-Ban Meta Gen 2. Needs a Mac with Xcode and a paid Apple Developer membership. Android phones are not built from Graspable yet. Run on my phone, above the preview of a Meta glasses app: Graspable builds the app with your Xcode, installs it on the iPhone and","u":"/changelog#v0.16.4"},{"k":"changelog","t":"Meta glasses, a first look","h":"Version 0.16.3","x":"Three templates for Ray-Ban Meta, Oakley Meta and Meta Ray-Ban Display glasses, and a glasses emulator in the preview. Tried so far on Ray-Ban Meta Gen 2 with an iPhone. Android and the Display glasses are not tried on a device yet. Meta glasses app: an Expo project on the Meta Wearables Device Access Toolkit. One screen per capability: registering with the Meta AI app, the camera stream with soun","u":"/changelog#v0.16.3"},{"k":"changelog","t":"Helpers, and interrupted runs that move with a project","h":"Version 0.16.1","x":"The Graspable agent can hand a self-contained part of a request to a helper with the same tools. A run that was interrupted no longer holds a project in place: it moves with the project and continues on the other side. The Graspable agent can hand a self-contained part of a request to a helper: a second conversation with the same tools in the same project. The helper's tool calls show in the run p","u":"/changelog#v0.16.1"},{"k":"changelog","t":"A run survives the app restarting","h":"Version 0.16.0","x":"The Graspable agent now writes down every step of a run as it happens. When the app updates itself, crashes or is quit while the agent works, the run continues from its last step as soon as you open the project again. A run of the Graspable agent that the app interrupted continues where it stopped: the conversation, the tool calls and the checks go on from the last step that was written down. It a","u":"/changelog#v0.16.0"},{"k":"changelog","t":"The agent is harder to talk into things","h":"Version 0.15.7","x":"A second security pass, this time on what text from outside can make the agent do: files in a forked project, results from connected services, what it remembers, and what it shows you. Text from outside the conversation that reads like instructions to the agent (a file in a project you continued from, a page or message a connected service returned, a plugin's answer) is marked as such before the a","u":"/changelog#v0.15.7"},{"k":"changelog","t":"A security review, and what came of it","h":"Version 0.15.6","x":"We went through Graspable with the OWASP checklists and known-vulnerability databases, fixed what we found, and gave every web template a security review skill. Every web template comes with a security review skill: ask the agent whether your app is safe to publish and it reviews the project against the OWASP Top 10. Continuing from an app through a link on a web page now asks you first: a project","u":"/changelog#v0.15.6"},{"k":"changelog","t":"Your published apps can use your connected services","h":"Version 0.15.5","x":"An app you publish can read your spreadsheet or post to your channel while it runs, through a grant you make for it. People who open the app sign in to nothing, and the app holds no key. Connectors for published apps: on graspable.dev, under Account, Published apps, Edit its page, grant an app one tool of a service you connected, with the values the app may not change (the spreadsheet, the folder,","u":"/changelog#v0.15.5"},{"k":"changelog","t":"React XR starts with a working scene, and Graspable for Linux","h":"Version 0.15.4","x":"The free template no longer opens on a single box: a new project is a small scene that already does the things most apps need. A new React XR project starts with a bench in front of you: things to pick up and throw, two buttons, a sign you can read, and a basket that counts what lands in it. It works with controllers, tracked hands and a mouse, in VR and in AR. Graspable for Linux on x64 computers","u":"/changelog#v0.15.4"},{"k":"changelog","t":"See what each template starts you with","h":"Version 0.15.3","x":"Every template page now shows the project you get when you choose it, before you change anything. Each template's picture is a new project from it, untouched, as it first opens: in the emulated headset for WebXR, in Unity's play mode, in the visionOS Simulator, or as the Unreal level on a flat screen. A new project from the XR Game template could open with an empty preview.","u":"/changelog#v0.15.3"},{"k":"changelog","t":"A page for every app, an account with sections, and a profile with your links","h":"Version 0.15.2","x":"Every published app has a page you fill in yourself, your account on the website is laid out in sections, and your profile lists where people find you. Every public app has its own page on graspable.dev with pictures, films, a longer text, who made it and where it runs. Add up to 8 pictures and 2 films from YouTube or Vimeo, choose the picture it is shown with, and say where feedback goes: Account","u":"/changelog#v0.15.2"},{"k":"changelog","t":"Your account on the website, and a fuller profile","h":"Version 0.15.1","x":"Everything about your account that does not need the app is now on graspable.dev, and your public profile can say more about you. Your account on graspable.dev/account: your plan and Credits, subscribing and buying Credits by card or with crypto, payments and invoices, your team, your published apps and your email preferences. Your profile on graspable.dev/profile: where you are, up to five links,","u":"/changelog#v0.15.1"},{"k":"changelog","t":"The agent on Credits, and three times the Credits for the first 100 on Builder and Pro","h":"Version 0.15.0","x":"You no longer need an AI account of your own to start, and the first 100 accounts on Builder or Pro get three times their plan's Credits for three months. The Graspable agent on Credits: with no AI account of your own, the agent runs on Graspable's and each request costs Credits by what it used, 10 Credits for every dollar the model costs. Under Settings, AI for the agent. Your own accounts stay f","u":"/changelog#v0.15.0"},{"k":"changelog","t":"A gallery of apps, and documentation you can search","h":"Version 0.14.0","x":"Every public app is now in one gallery with search, categories and tags, and the documentation has three parts and a search over all of them. The gallery on graspable.dev/made: every public app, searched by name, tag or maker, filtered by category, ordered by trending, most viewed, most liked or newest. Give your public app a category and up to five tags under Settings, Published apps, so people f","u":"/changelog#v0.14.0"},{"k":"changelog","t":"Captures, and a picture for every project","h":"Version 0.13.0","x":"Take a picture of what you see in the preview, keep it, share it, or make it the picture your project is shown with. Capture, above the preview: a picture from where you stand and look, always 1280 × 720. Look at it first, then keep it, download it, or take it again. Use a capture as the project's picture: lists and your published app show it. Until you choose one, it is the app as a person first ","u":"/changelog#v0.13.0"},{"k":"changelog","t":"Sign in on the website, and a page of your own","h":"Version 0.12.1","x":"Community now works on graspable.dev too, and everyone who publishes gets a page worth sharing. Sign in on graspable.dev with the account you use in the app, and like apps, save them to collections and follow people right there. Your page at graspable.dev/@yourhandle: your apps with their pictures, views and likes, your followers and your public collections. Say a few words about yourself, add a l","u":"/changelog#v0.12.1"},{"k":"changelog","t":"Community","h":"Version 0.12.0","x":"Public apps can now be found, liked and collected, and you can follow the people who make them. Every published app counts its views. Community in the app: public apps by most viewed, most liked, newest, and from the people you follow. Open one, or continue from it. Like an app, follow who made it, and save apps to collections. A collection has a page of its own to share. Every published app count","u":"/changelog#v0.12.0"},{"k":"changelog","t":"Pay with crypto","h":"Version 0.11.0","x":"Credits, AI for the agent and what the agent buys from outside services can now be paid with crypto. All of it is young: it is marked experimental in the app, so start with small amounts. Buy Credits with Bitcoin, Ether, Solana or USDC on Base, from your own wallet: the app shows the amount, the address and a QR code. The same prices as by card, no card and no identity check. Credits are added onc","u":"/changelog#v0.11.0"},{"k":"changelog","t":"The agent knows what you asked before","h":"Version 0.10.5","x":"Every request now starts with what the last ones were, so a follow-up such as \"make it bigger\" needs less looking around. With the Graspable agent, a quick follow-up simply goes on with the same conversation. Each request begins with a short account of your last five: what you asked, which files changed and how it ended. It is made on your computer, for every agent. Graspable agent: when you ask a","u":"/changelog#v0.10.5"},{"k":"changelog","t":"Native XR: Unity, Unreal Engine and Apple Vision Pro","h":"Version 0.10.0","x":"Graspable now builds native headset apps as well as web ones. This release gathers what the last weeks added: three native engines, an asset browser, skills, and an agent that finishes sooner. Unity templates for Meta Quest, Android XR and XREAL: a mixed-reality kit with hands and controllers, real materials, a simulated room in the Editor, and an Android app built for the headset on USB. An Unrea","u":"/changelog#v0.10.0"},{"k":"changelog","t":"Shorter requests on every agent","h":"Version 0.9.61","x":"The agent checks its work in fewer, quicker steps. The same request that took Claude Code 12 to 27 minutes in our tests now takes about 10, and Codex went from about 10 minutes to under 6. A check in the emulated headset builds the project and looks at it at the same time: about 6 seconds instead of 12 on a changed project. The agent is asked to check finished steps, not every edit, and a check sa","u":"/changelog#v0.9.61"},{"k":"changelog","t":"Faster checks","h":"Version 0.9.60","x":"The agent spends less time checking its work: a project that has not changed since a check passed is not checked again, and an Unreal check takes about a third less time. When the agent has just checked the project and nothing changed since, the check at the end of the request is not repeated. This holds for every kind of project. Unreal Engine: a level that has not changed is not built again, and","u":"/changelog#v0.9.60"},{"k":"changelog","t":"An asset browser, skills per project, and C++ in Unreal","h":"Version 0.9.59","x":"A project's models, images and sounds are now listed beside its files, with a 3D view of each model. Skills can be switched per project, and Unreal projects can have C++. Assets: a tab beside Files lists the project's models, images, sounds, videos and engine assets by folder. A model is drawn in 3D with its textures, its triangle count and its size; images, sound and video are shown and played. A","u":"/changelog#v0.9.59"},{"k":"changelog","t":"Unreal scenes that look like studio work","h":"Version 0.9.58","x":"The Unreal Engine template now builds with scanned furniture and props, real daylight and Unreal's high renderer, keeps a headset budget, and comes with skills for conventions, performance and look. Unreal Engine: scanned models, photographed surfaces and skies from Poly Haven by name, at real size; glass, glowing materials, spot and area lights; daylight in real units with a camera-style exposure","u":"/changelog#v0.9.58"},{"k":"changelog","t":"Steadier Unreal pictures, and an example with a Blueprint","h":"Version 0.9.57","x":"The picture of an Unreal Engine 5.8 project waits for its materials more carefully, and Made in Graspable has a start-up station whose lever switches a lamp. Unreal Engine 5.8: the agent's Editor is closed before the picture is taken, the picture is repeated while materials are still being prepared, and the agent is told that a picture of grey checkers means \"check again\". On a very busy computer ","u":"/changelog#v0.9.57"},{"k":"changelog","t":"A plainer download, and Unreal pictures that show the materials","h":"Version 0.9.56","x":"The headset app download on a published app is one button with what matters beside it. Unreal Engine 5.8 projects get a correct picture after the agent used the Editor. A published native app shows its headset app as one button with its size; the checksum is folded away under it. Unreal Engine 5.8: the picture the agent checks its work with showed grey checkers instead of materials after the agent","u":"/changelog#v0.9.56"},{"k":"changelog","t":"Blueprints in Unreal, headset apps as downloads, and native XR on the Pro plan","h":"Version 0.9.54","x":"With Unreal Engine 5.8 the agent can write Blueprints in your own Unreal Editor (experimental). A published native app can offer its headset app as a download and link to its store pages. Native XR now comes with the Pro plan. Experimental, Unreal Engine 5.8 and later: the agent reads, writes and compiles Blueprints through the Editor's own tools, in an Editor that runs without a window. The level","u":"/changelog#v0.9.54"},{"k":"changelog","t":"Hands, controllers and things to do, in Unity and Unreal","h":"Version 0.9.53","x":"Native scenes are no longer only to look at: full hands and controllers in Unity with levers, drawers and panels, and Unreal scenes with a rig, things to grab and real joints. Unity: the templates use Unity's rig for hands and controllers together: visible hands, fingertip poke, pinch near and far, controller rays, grab, teleport, turn, climb and gaze. Unity: mechanisms that a hand moves (levers, ","u":"/changelog#v0.9.53"},{"k":"changelog","t":"Unreal Engine, and the native examples rebuilt","h":"Version 0.9.52","x":"A first Unreal Engine template whose level is a Python script, and the three Unity examples rebuilt with real materials in a simulated workshop. Unreal Engine XR, a template for Unreal Engine 5 with OpenXR. Its level is a Python script that your own Unreal Editor builds, so the agent can change the scene and you can read what changed. The agent sees a picture of the game. It builds scenes, materia","u":"/changelog#v0.9.52"},{"k":"changelog","t":"Real materials, a room in the Editor, one app for every device","h":"Version 0.9.51","x":"Native projects look like what they are made of, Unity play mode runs in a simulated room, the agent works inside the Unity Editor, and a new template makes one WebXR app for every headset. Cross-platform XR, a template for one WebXR app that adapts to the device: in your own room where the headset shows it, in a photographed room where it does not, with a panel built with uikit, photographed mode","u":"/changelog#v0.9.51"},{"k":"changelog","t":"Native examples in Made in Graspable","h":"Version 0.9.50","x":"Five new examples in Made in Graspable for headsets and glasses, each built from one request; native projects can now be published and continued from. Five examples of the kind companies commission: a guided pump shutdown for Meta Quest, a gearbox design review for Android XR, pick by vision for XREAL glasses, a shelving configurator for Apple Vision Pro, and a work-order console as a panel app. E","u":"/changelog#v0.9.50"},{"k":"changelog","t":"The Vision Pro check sees the space","h":"Version 0.9.49","x":"For an Apple Vision Pro project, the agent's check and the preview now show the immersive space together with the window. Apple Vision Pro: the check opens the app's immersive space in the visionOS Simulator, so the picture the agent and you see has the window and the things in the space, not the window alone.","u":"/changelog#v0.9.49"},{"k":"changelog","t":"Mixed reality in Unity, Apple Vision Pro, panel apps","h":"Version 0.9.48","x":"Graspable's own Unity project for Quest, Android XR and XREAL with passthrough and real surfaces, a Swift template for Apple Vision Pro run in the visionOS Simulator, an Expo panel app, and Build for headset. The Unity templates are now Graspable's own project, not Unity's sample: a mixed-reality start scene (the real room shows through, a ball to grab, a button to press, a marker you place on rea","u":"/changelog#v0.9.48"},{"k":"changelog","t":"Native headset apps with Unity","h":"Version 0.9.47","x":"A Unity for Quest template: the agent writes C# and scenes, a windowless Unity Editor compiles them and renders a frame of play mode. Unity for Quest, a template for native Meta Quest and Meta VR Glasses apps (Unity 6, OpenXR, XR Interaction Toolkit, hands). Builder plan and higher; needs Unity 6 and the Unity CLI on your computer. For a Unity project the agent's check compiles the scripts with th","u":"/changelog#v0.9.47"},{"k":"changelog","t":"Meta VR Glasses in the emulated headset","h":"Version 0.9.46","x":"Try a scene the way Meta VR Glasses will show it: gaze to aim, pinch to select, a narrower view. The emulated headset in the preview can be Meta VR Glasses: choose it in the bar above the preview. Recordings made with it replay with it, and the agent can check an app on the glasses with verify_app. The agent is told what the glasses need: every interaction through a ray and one select, targets big","u":"/changelog#v0.9.46"},{"k":"changelog","t":"A page for every example and template","h":"Version 0.9.45","x":"The site has a page for each example with the request that built it, and for each template. On the site: every example in Made in Graspable has its own page with the request that built it, word for word, and every template has a page with requests to try. A published app now shows its own name and description when its link is shared, instead of the template's name. Four optional welcome questions ","u":"/changelog#v0.9.45"},{"k":"changelog","t":"Payments and invoices in the app","h":"Version 0.9.42","x":"See every payment with its invoice in Settings, and a fix for models that were just added. Payments and invoices: Settings → Plan and Credits lists every payment, each with its invoice, and a button to change your payment method or billing details. Opening billing could fail with “Failed to create portal session”. It now works for every account that has paid, including one that only bought Credits","u":"/changelog#v0.9.42"},{"k":"changelog","t":"Choose the services a project uses","h":"Version 0.9.41","x":"Add a connected service to a project right next to the box where you write to the agent. Connect apps, next to the prompt: add the connected services a project should use, and remove them again. The agent gets a service's tools only in the projects you added it to. After connecting a service, the page in your browser could say the account was not connected when it was. It now looks at the account ","u":"/changelog#v0.9.41"},{"k":"changelog","t":"Your own cloud on more plans","h":"Version 0.9.37","x":"Cloud Agents in your own E2B account now come with Builder, and on your own server with Pro. Cloud Agents in your own E2B account come with the Builder plan and higher; on a server of your own, with Pro and higher. Both were Studio only. Report an app: anyone can tell us about a published app that is deceptive, harmful or copies their work, at graspable.dev/report. A person on the team reads every","u":"/changelog#v0.9.37"},{"k":"changelog","t":"Every published app at its own address","h":"Version 0.9.36","x":"A published app now lives at its own address on graspable.cloud, apart from every other app. Every published app has an address of its own, made from its name: an app named hoops is at hoops.graspable.cloud. Links to an app's page on graspable.dev keep working. New Multi-user projects meet through realtime.graspable.dev. The Pipedream connector that needed a Pipedream project of your own has been ","u":"/changelog#v0.9.36"},{"k":"changelog","t":"Your own splats, and longer runs with Claude Code","h":"Version 0.9.35","x":"Bring a Gaussian splat you captured yourself into the world templates, and let Claude Code work through large builds. Your own splat as a world: in a project made from Splat World, World Explorer or World Game, choose Worlds in the header and pick a .spz, .ply, .splat, .ksplat or .sog file (up to 500 MB). The agent switches the app to it and sets its size and floor when you tell it what is off. El","u":"/changelog#v0.9.35"},{"k":"changelog","t":"Publish your apps, continue from anyone's, and a history of every project","h":"Version 0.9.32","x":"Every project can be a page on graspable.dev at your own handle. Public apps can be continued by anyone in Graspable, and each project keeps a history you can go back to. Publish: one button puts your app on the web at an address of its own, hosted by Graspable. Builder and higher can keep an app private, behind a link with a key. Handles: choose yours in Settings → Graspable account. Your page at","u":"/changelog#v0.9.32"},{"k":"changelog","t":"Generated worlds load in the templates","h":"Version 0.9.31","x":"A fix for worlds generated from inside a project. A newly generated world was saved but did not appear, because the templates could not find its files.","u":"/changelog#v0.9.31"},{"k":"changelog","t":"Seven premium templates, and worlds from World Labs","h":"Version 0.9.30","x":"Ask for a place and stand in it. Six new premium templates start you much further along. Generated worlds: ask for a place in plain words and the agent generates it with World Labs and puts it in your project. 2 Credits for a draft, 11 for a standard world, 21 for a large one, or use your own World Labs key. World Explorer: walk through a generated place as if it were solid, with info points and a","u":"/changelog#v0.9.30"},{"k":"changelog","t":"Two fixes for checks","h":"Version 0.9.29","x":"Recorded XR tests replay again, and Codex can repair a failed check. Replaying a recorded XR test could fail with \"Invalid hook call\" in projects created with 0.9.27 or 0.9.28. With the Codex CLI, a run whose check failed stopped instead of trying to repair it.","u":"/changelog#v0.9.29"},{"k":"changelog","t":"Plugins, skills, and Graspable for other AI agents","h":"Version 0.9.28","x":"Give the agent tools from other services, teach it how you work, and let your own AI agent build with Graspable. Plugins and connectors give the agent tools from other services and programs. Any MCP server can be added, and a Pipedream connector is included. Skills: a folder with your instructions for one kind of task. The agent reads a skill when a task calls for it. Plugins and skills work the s","u":"/changelog#v0.9.28"},{"k":"changelog","t":"Nothing to install first, and updates in one click","h":"Version 0.9.27","x":"Graspable no longer needs Node.js or a browser set up beforehand, and it updates itself. What runs your projects is built into the app. You do not need Node.js. If your computer has no Chrome, Edge or Chromium, the app downloads a browser once to check that your scenes load. Settings → Components shows both. Under Advanced you can use your own Node.js or browser instead. When a new version is out,","u":"/changelog#v0.9.27"},{"k":"changelog","t":"Choose the AI per project, buy any amount of Credits","h":"Version 0.9.26","x":"Each project can use a different AI, and cloud projects pick an API key by themselves. A list above the prompt chooses the AI for that project. Settings keeps the default. A cloud project now uses a saved API key by itself when your default is a ChatGPT or Claude subscription, instead of refusing to run. Buy any amount of Credits from 50, next to the three packs. Settings → Plan shows buying Credi","u":"/changelog#v0.9.26"},{"k":"changelog","t":"Fast decisions for everyone","h":"Version 0.9.25","x":"System One, the small model that makes quick judgements during a run, is now included when you are signed in. System One works without any setup when you are signed in to your Graspable account. You can still add your own model. It now also leaves out saved notes that are not about your request, keeps one-off details out of memory, and stops a run that keeps repeating the same failed fix. You see ","u":"/changelog#v0.9.25"},{"k":"changelog","t":"Cloud projects keep working after an app update","h":"Version 0.9.24","x":"The app and a Cloud Agent no longer have to be the same version. After updating the app, a cloud project could refuse to open with \"They must match\". It opens now. Update the cloud agent: one click in the project cloud menu moves the project to a Cloud Agent of your app version. With a paused Cloud Agent that could not wake, clicking Preview made the whole project page stop responding. The page no","u":"/changelog#v0.9.24"},{"k":"changelog","t":"A sharper handover after a move","h":"Version 0.9.23","x":"With a System One model set up, the account of earlier work is picked passage by passage for your new request. After a project moves, a fast decision model can choose which parts of the earlier conversation your next request needs. It takes about a third of a second and uses no Credits. The payment service can now add new kinds of notices without causing errors on our side.","u":"/changelog#v0.9.23"},{"k":"changelog","t":"The agent picks up where it left off after a move","h":"Version 0.9.22","x":"When a project moves to the cloud or back, the next run starts with a short account of the earlier work. The first run after a move knows what was asked before, which files changed and how each run ended. With Fast mode, that account is cut down to what matters for your new request, so the agent rarely needs to re-read the project. Undoing a run is now shown in the conversation and taken into acco","u":"/changelog#v0.9.22"},{"k":"changelog","t":"One conversation per project","h":"Version 0.9.21","x":"The run sidebar now reads as one thread, and it stays with the project when it moves to the cloud and back. Earlier requests and answers of a project are shown above the current run, each with a link to replay that run. After moving a project to the cloud, the run history could look empty until you touched the timeline. It now reloads by itself. Replaying an empty XR recording showed \"wrong record","u":"/changelog#v0.9.21"},{"k":"changelog","t":"Teams, Credit packs and new prices","h":"Version 0.9.20","x":"Work on a cloud project together, buy Credits when you need them, and pay less for Builder and Pro. Teams on the Studio plan: invite people by email, share cloud projects, and spend from one pool of Credits. Settings → Team shows who used what. Credit packs: buy 100, 300 or 1,000 Credits at any time. They are used after your plan Credits and last 12 months. Overage: let work continue when Credits ","u":"/changelog#v0.9.20"},{"k":"changelog","t":"Fast mode","h":"Version 0.9.19","x":"On the Pro plan the agent can edit, search and read build output faster. Fast edit: the agent writes only the lines that change, and the file is updated in about a second. Code search: the agent finds code by what it does, not only by exact words. Long build output is shortened to the lines that matter before the agent reads it. When a fix keeps failing the same way, the run stops instead of spend","u":"/changelog#v0.9.19"},{"k":"changelog","t":"Plans that match the product, and Cloud Agents in Graspable cloud","h":"Version 0.9.18","x":"Building on your computer stays free. Paid plans add Cloud Agents, premium templates and Credits. Cloud Agents now run in Graspable cloud with nothing to set up. They use Credits: 1 Credit for every 10 minutes. Builder, Pro and Studio plans include Credits each month. Settings → Plan shows what you have left. Community templates are free for everyone. Premium templates come with Builder and higher","u":"/changelog#v0.9.18"},{"k":"changelog","t":"Cloud agents","h":"Version 0.9.16","x":"A project can now run in the cloud instead of on your computer, and come back whenever you like. Move a project to the cloud from its header. The agent, its commands and the live preview run there and keep going while your computer sleeps. Move it back with one click. Your files, run history and memory travel both ways. An idle cloud project pauses by itself and wakes when you use it.","u":"/changelog#v0.9.16"},{"k":"changelog","t":"A preview you can always see","h":"Version 0.9.15","x":"Fixes from the first long sessions in the installed app. The live preview stayed empty in the installed app. It shows now. A headset test recorded while the scene was broken could make every later run fail. Empty recordings are no longer saved. A rating you give while offline or signed out is kept and sent later. New scenes fill the whole window and keep their controls inside the scene, where a he","u":"/changelog#v0.9.15"},{"k":"changelog","t":"New models, no update needed","h":"Version 0.9.14","x":"When Claude Code, Codex or your provider adds a model, Graspable picks it up. A model picker for Claude Code and the Codex CLI, filled from the tools themselves. “Refresh model list” in Settings checks for new models on demand.","u":"/changelog#v0.9.14"},{"k":"changelog","t":"Smoother sign-in","h":"Version 0.9.12","x":"Signing in is clearer and the app stops asking for your password. macOS no longer asks for your keychain password again and again. After signing in with ChatGPT, Settings now says so plainly.","u":"/changelog#v0.9.12"},{"k":"changelog","t":"Memory that helps","h":"Version 0.9.10","x":"The agent remembers what matters and brings it up when it is relevant. For each request the agent recalls the few saved notes that fit it. The run shows which ones. Open any memory to correct it, pin it or forget it. What you tell the agent always outranks what it worked out by itself.","u":"/changelog#v0.9.10"},{"k":"changelog","t":"Sign in from inside the app","h":"Version 0.9.9","x":"No terminal, no copying keys around. One place in Settings to choose what powers the agent: your ChatGPT subscription, Claude Code with your Claude subscription, or an API key. Graspable can install Claude Code and sign you in to it. Signing in to your Graspable account with GitHub or Google now returns to the app.","u":"/changelog#v0.9.9"},{"k":"changelog","t":"Tell us how it went","h":"Version 0.9.5","x":"Every answer from the agent can be rated. Thumbs up, thumbs down and copy under each response. A thumbs down can say why.","u":"/changelog#v0.9.5"},{"k":"changelog","t":"See the scene, find the code","h":"Version 0.9.4","x":"Tools for working on a scene directly. A Scene panel lists everything in your scene. Click an object to jump to the code that makes it. Try your scene on a real headset over Wi-Fi by scanning a QR code, or over USB. Before you approve a change, see exactly what will change.","u":"/changelog#v0.9.4"},{"k":"changelog","t":"The desktop studio","h":"Version 0.9.0","x":"Graspable becomes a desktop app that builds real projects on your computer. Describe a scene and watch the agent build it in a project folder you own. A run only finishes when the project builds and loads in a browser without errors. An emulated headset in a scanned room. Record a session and it is replayed as a test after every change. Undo a whole run with one click. Drag any number in the code ","u":"/changelog#v0.9.0"},{"k":"blog","t":"Lessons students can step into","x":"Three lessons made in Graspable that students open from a link: a solar system they pick up, chime bars they play with their hands, and a guided tour through a generated greenhouse. A diagram on a slide asks a student to imagine something. A model they can pick up, turn and open asks much less of them. XR has been good at this for years, but making a lesson meant learning a game engine first. In G","s":"Three lessons made in Graspable that students open from a link: a solar system they pick up, chime bars they play with their hands, and a guided tour through a generated greenhouse. A diagram on a slide asks a student to imagine something. A model they can pick up, turn and open asks much less of them. XR has been good at this for years, but making a lesson meant learning a game engine first. In Graspable a teacher or course designer describes the lesson, an agent builds it as a web project, checks that it runs, and the class opens it from a link in a headset or a browser. These three lessons were each made that way, and each can be opened and continued from.","u":"/blog/xr-lessons-students-can-step-into"},{"k":"blog","t":"Lessons students can step into","h":"A solar system you can pick up","x":"Planets circle the Sun on brass rings in a quiet planetarium. Grab one and its lesson card opens beside it: diameter, length of a day and a year, distance from the Sun. Let go and it returns to its orbit. Built from the React XR template, which is free. Open it The request that built it describes the room, the rings and what each card says. Changing the lesson is another request: add the moons, tr","s":"Planets circle the Sun on brass rings in a quiet planetarium. Grab one and its lesson card opens beside it: diameter, length of a day and a year, distance from the Sun. Let go and it returns to its orbit. Built from the React XR template, which is free. Open it The request that built it describes the room, the rings and what each card says. Changing the lesson is another request: add the moons, translate the cards, or end with three questions.","u":"/blog/xr-lessons-students-can-step-into#a-solar-system-you-can-pick-up"},{"k":"blog","t":"Lessons students can step into","h":"Chime bars for bare hands","x":"Eight coloured bars on a walnut frame. Touch one with a fingertip and it dips, glows and rings its note. A small panel above holds a volume slider and a button that plays a tune. With hand tracking there is nothing to hold, which makes it a good first lesson for people who have never worn a headset. Built from the Hands and Spatial UI template. Open it","u":"/blog/xr-lessons-students-can-step-into#chime-bars-for-bare-hands"},{"k":"blog","t":"Lessons students can step into","h":"A guided tour with stops","x":"A five-stop walk through a generated Victorian greenhouse at dusk. Each stop has a plate with two lines of text, and Next takes the group on. The same structure carries a field trip, a tour of a historic building or a walk through a process. Built from the World Explorer template, in a world generated from a sentence. Open it","u":"/blog/xr-lessons-students-can-step-into#a-guided-tour-with-stops"},{"k":"blog","t":"Lessons students can step into","h":"Using this in a class","x":"- One link for everyone. Publish a lesson and it has its own address. It opens in Meta Quest and other headsets through the browser, and on school computers without a headset. - Try it without a headset first. The preview in Graspable has an emulated headset, so you can walk through the lesson at your desk before class. - It stays yours. The lesson is a project on your computer. You can change it,","s":"- One link for everyone. Publish a lesson and it has its own address. It opens in Meta Quest and other headsets through the browser, and on school computers without a headset. - Try it without a headset first. The preview in Graspable has an emulated headset, so you can walk through the lesson at your desk before class. - It stays yours. The lesson is a project on your computer. You can change it, copy it for another class, or hand it to a colleague. - Free to start. Building with your own AI account costs nothing, and the React XR template is free. The education page collects these and the templates that suit lessons.","u":"/blog/xr-lessons-students-can-step-into#using-this-in-a-class"},{"k":"blog","t":"XR training for industry, by describing it","x":"Five working examples of XR training and operations apps made in Graspable: a guided pump shutdown, a machine start-up console, a gearbox design review, a technician work-order panel and pick-by-vision for a warehouse. Most industrial XR projects start with the same request: show a person how to do a procedure at the machine, or let them practise it before they touch the real one. Building that us","s":"Five working examples of XR training and operations apps made in Graspable: a guided pump shutdown, a machine start-up console, a gearbox design review, a technician work-order panel and pick-by-vision for a warehouse. Most industrial XR projects start with the same request: show a person how to do a procedure at the machine, or let them practise it before they touch the real one. Building that used to mean a specialist studio and months of work. In Graspable you describe the procedure, an agent builds it as a real Unity, Unreal or web project on your computer, checks that it runs, and you try it in the editor's play mode or a headset. These five examples were each made that way. Every one is published with its source, so you can open it and continue from it in your own copy.","u":"/blog/xr-training-for-industry"},{"k":"blog","t":"XR training for industry, by describing it","h":"A guided shutdown of a pump unit","x":"A five-step maintenance procedure in mixed reality. The step card says what to do, the part to act on glows, and the unit answers: the motor slows when it is switched off, the gauge drops when the valve is closed. The next step waits until the current one is done. Built from the Unity for Quest template. Open it This is the shape of most work instructions: a fixed order, one part at a time, and a ","s":"A five-step maintenance procedure in mixed reality. The step card says what to do, the part to act on glows, and the unit answers: the motor slows when it is switched off, the gauge drops when the valve is closed. The next step waits until the current one is done. Built from the Unity for Quest template. Open it This is the shape of most work instructions: a fixed order, one part at a time, and a clear sign that a step is done. The same structure works for a lockout procedure, a filter change or a first inspection.","u":"/blog/xr-training-for-industry#a-guided-shutdown-of-a-pump-unit"},{"k":"blog","t":"XR training for industry, by describing it","h":"A machine start-up station","x":"A plant-room console for training new operators, built in Unreal Engine. Pull the main power lever down and the work lamp comes on and the sign changes to read POWER ON. It is small on purpose: one action and an answer from the room is often the first thing a trainee needs to practise. Built from the Unreal Engine XR template. Open it","u":"/blog/xr-training-for-industry#a-machine-start-up-station"},{"k":"blog","t":"XR training for industry, by describing it","h":"A design review of a gearbox","x":"A gearbox at real size that comes apart. Explode it, cut it open, select a part to read its material and mass, and pin notes to it. Engineering teams use this kind of tool to look at an assembly together before anything is machined, and the same project is a good onboarding model for people who will service it. Built from the Unity for Android XR template. Open it","u":"/blog/xr-training-for-industry#a-design-review-of-a-gearbox"},{"k":"blog","t":"XR training for industry, by describing it","h":"A work-order console for technicians","x":"A panel that floats beside the machine: pick a work order, answer its checklist with Pass, Fail or N/A, watch the asset's readings, and sign off. The readings in the example are simulated and labelled so. In a real deployment they would come from your own systems through an API, which the agent can connect when you give it the address. Built from the Expo panel app template. Open it","u":"/blog/xr-training-for-industry#a-work-order-console-for-technicians"},{"k":"blog","t":"XR training for industry, by describing it","h":"Pick by vision for a warehouse","x":"On AR glasses the view is narrow, so the next pick stays on a strip in the middle of it, the right bin lights up, and a pinch confirms the pick and moves to the next one. Built from the Unity for XREAL template. Open it","u":"/blog/xr-training-for-industry#pick-by-vision-for-a-warehouse"},{"k":"blog","t":"XR training for industry, by describing it","h":"What this means for a training team","x":"- You own the project. It is a normal Unity, Unreal or web project on your computer, with its history. Your developers can open it in the engine and change anything by hand. - Changes are a sentence. \"Add a step that checks the pressure is below 1 bar before the valve opens\" is a request, not a ticket. - It is checked before you see it. A run ends only when the project builds and runs. For Unity a","s":"- You own the project. It is a normal Unity, Unreal or web project on your computer, with its history. Your developers can open it in the engine and change anything by hand. - Changes are a sentence. \"Add a step that checks the pressure is below 1 bar before the valve opens\" is a request, not a ticket. - It is checked before you see it. A run ends only when the project builds and runs. For Unity and Unreal that means the engine's own build and a picture from play mode. - It runs where your people are. Meta Quest, Android XR and XREAL glasses through Unity, high-fidelity rooms through Unreal, and anything with a browser through WebXR. Native engine templates come with the Pro plan, and Graspable uses the Unity or Unreal Engine already installed on your computer. If you want to talk about a training programme, the industrial training page has the details and a way to reach us.","u":"/blog/xr-training-for-industry#what-this-means-for-a-training-team"},{"k":"blog","t":"Build inside XR by talking","x":"Graspable now has voice. Speak a request instead of typing it, hold a conversation with the agent and hear its answers, and keep talking after you put the headset on. A headset has no keyboard worth using. Until now that meant a loop: look at the scene in the headset, take the headset off, type what should change, put it back on. Graspable 0.21 closes that loop with voice. You can speak a request ","s":"Graspable now has voice. Speak a request instead of typing it, hold a conversation with the agent and hear its answers, and keep talking after you put the headset on. A headset has no keyboard worth using. Until now that meant a loop: look at the scene in the headset, take the headset off, type what should change, put it back on. Graspable 0.21 closes that loop with voice. You can speak a request at your desk, hold a conversation with the agent while it works, and carry on that conversation from inside the scene. A reading corner built by conversation, filmed in the emulated headset. The film is cut for length: how long a change takes depends on the model you use.","u":"/blog/talk-to-the-agent"},{"k":"blog","t":"Build inside XR by talking","h":"Speak a request","x":"A microphone now sits beside the prompt. Press it, say what you want, and press it again or hit Enter. Your words appear in the prompt as text, so you can fix a word before you send it. Escape drops the recording. While it listens, the microphone becomes a stop button. Graspable hears you only between the two presses. This is the small version of voice, and for many requests it is enough. Describi","s":"A microphone now sits beside the prompt. Press it, say what you want, and press it again or hit Enter. Your words appear in the prompt as text, so you can fix a word before you send it. Escape drops the recording. While it listens, the microphone becomes a stop button. Graspable hears you only between the two presses. This is the small version of voice, and for many requests it is enough. Describing where a shelf should go is faster out loud than typed, and you still get to read the request before it becomes work.","u":"/blog/talk-to-the-agent#speak-a-request"},{"k":"blog","t":"Build inside XR by talking","h":"Talk it through","x":"The button with the sound waves, under the microphone, starts a conversation. The microphone stays open, and what you say is sent when you stop speaking. If the agent is idle, your words become a new request. If it is in the middle of a run, they reach it as a note that steers the work, so \"make the lamp taller\" can land while the lamp is still being built. The agent answers in a few short sentenc","s":"The button with the sound waves, under the microphone, starts a conversation. The microphone stays open, and what you say is sent when you stop speaking. If the agent is idle, your words become a new request. If it is in the middle of a run, they reach it as a note that steers the work, so \"make the lamp taller\" can land while the lamp is still being built. The agent answers in a few short sentences, read aloud. Speak while it is reading and it stops, the way you would interrupt a person. When a step needs your approval, the agent says so, and you approve or decline on the screen. The bar above the prompt says what the conversation is doing: listening, hearing you, sending, or answering. Every agent message also has a Read aloud button beside Copy, for when you want to hear one answer without starting a conversation.","u":"/blog/talk-to-the-agent#talk-it-through"},{"k":"blog","t":"Build inside XR by talking","h":"Keep talking in the headset","x":"The same conversation works in VR and AR. A Talk to the agent button sits low in front of you, out of the way of the scene and one glance down. Press it and speak. The Talk button shows what the conversation is doing, so you know when the agent has heard you and when it is answering. In the preview on your computer, the emulated headset uses the computer's microphone. On a real headset, open Heads","s":"The same conversation works in VR and AR. A Talk to the agent button sits low in front of you, out of the way of the scene and one glance down. Press it and speak. The Talk button shows what the conversation is doing, so you know when the agent has heard you and when it is answering. In the preview on your computer, the emulated headset uses the computer's microphone. On a real headset, open Headset , share the preview on Wi-Fi, choose Talk in the headset and scan the code. The headset then listens with its own microphone and plays the agent's answers in your ears. The conversation also shows in the app on your computer, which is where you approve steps. The headset sends its recordings to Graspable on your computer and nowhere else. It never holds your ElevenLabs key or your Graspable sign-in. This is the part that changes how building feels. You stand in the room you are making, say what is missing, and watch it arrive where you are looking. The scene in the film gives each new thing a pop and a few sparkles when it appears. That is ordinary scene code, and you can ask the agent for the same in your own project. 'Shelves on the left wall?' 'Perfect.'","u":"/blog/talk-to-the-agent#keep-talking-in-the-headset"},{"k":"blog","t":"Build inside XR by talking","h":"Who hears you","x":"ElevenLabs turns your speech into text and the agent's answers into speech. Under Settings, Voice you choose how that is paid for: - Graspable, on Credits. With your Graspable account, about ten minutes of speech are free every day. Beyond that, voice uses Credits by what it cost. - Your own ElevenLabs key. ElevenLabs bills your account directly. The key stays in your computer's keychain and is se","s":"ElevenLabs turns your speech into text and the agent's answers into speech. Under Settings, Voice you choose how that is paid for: - Graspable, on Credits. With your Graspable account, about ten minutes of speech are free every day. Beyond that, voice uses Credits by what it cost. - Your own ElevenLabs key. ElevenLabs bills your account directly. The key stays in your computer's keychain and is sent only to ElevenLabs. Graspable keeps no recording and no text of what you said. It counts the seconds turned into text and the characters read aloud that day, and that is all. Speech to text works out the language you speak by itself. If short requests confuse it, set your language in the same place. You also pick the agent's voice there, and Hear it plays a sample.","u":"/blog/talk-to-the-agent#who-hears-you"},{"k":"blog","t":"Build inside XR by talking","h":"Try it","x":"Voice is in Graspable 0.21.3 for Windows, macOS and Linux. Update the app, open Settings, Voice , and press the microphone. The details are in the voice documentation, and Preview and headset covers opening a project on a real headset.","u":"/blog/talk-to-the-agent#try-it"},{"k":"blog","t":"Design: one editor for the panels in your 3D scene","x":"Graspable 0.21 brings Design, an editor for the 2D interfaces inside a 3D scene. Every change is written into your code, the agent works on what you select, and one panel can be written for Apple Vision Pro, Unity headsets and Meta Ray-Ban Display. Most of what people do in a headset app, they do on a panel: a menu, a mixer, a settings page, a row of tools. Until now, every change to one in Graspa","s":"Graspable 0.21 brings Design, an editor for the 2D interfaces inside a 3D scene. Every change is written into your code, the agent works on what you select, and one panel can be written for Apple Vision Pro, Unity headsets and Meta Ray-Ban Display. Most of what people do in a headset app, they do on a panel: a menu, a mixer, a settings page, a row of tools. Until now, every change to one in Graspable went through the agent or the code. Design is an editor for those panels. It shows each one flat and straight on, like a screen in an interface builder, and writes every change into the code that places it, so the agent, the 3D view and the headset all see the same interface. Design in Graspable 0.21: drag, space, spring, theme, check, export, and the same panel in the emulated headset.","u":"/blog/design-panels-for-headsets-phones-and-glasses"},{"k":"blog","t":"Design: one editor for the panels in your 3D scene","h":"Edit by hand, or ask","x":"Click a part in the flat view or in the list beside it, and the inspector shows what that kind of part has: a button's label, size and colours, a slider's range and step, a panel's padding. Drag the handles on a selected part to size it. Select a row, column or panel, and the space between its parts shows in pink; drag it to change the gap, or drag a panel's edges to change its padding. The previe","s":"Click a part in the flat view or in the list beside it, and the inspector shows what that kind of part has: a button's label, size and colours, a slider's range and step, a panel's padding. Drag the handles on a selected part to size it. Select a row, column or panel, and the space between its parts shows in pink; drag it to change the gap, or drag a panel's edges to change its padding. The preview changes as you drag, and the whole drag is one step to undo. The selected part also goes with your next message, so \"make this bigger\" or \"add a volume slider under this\" is about that part. The agent's changes show in the flat view as it makes them. A selected row shows the space between its parts. Drag it to change the gap. Parts move by drag in the list: onto the edge of another part to go before or after it, onto a row, column or panel to go inside. The keys you expect work too: undo and redo, delete, duplicate, and the arrow keys to move a panel in the room.","u":"/blog/design-panels-for-headsets-phones-and-glasses#edit-by-hand-or-ask"},{"k":"blog","t":"Design: one editor for the panels in your 3D scene","h":"Motion with springs","x":"Buttons press in, toggles switch and sliders grow with springs, and a panel comes in with a short movement when it first appears. A spring keeps its speed when its target changes, so a press that is let go half way turns round smoothly instead of jumping. The Motion section of the inspector picks one of four springs, each set by how long it takes and how much it overshoots: Quick, Gentle, Bouncy a","s":"Buttons press in, toggles switch and sliders grow with springs, and a panel comes in with a short movement when it first appears. A spring keeps its speed when its target changes, so a press that is let go half way turns round smoothly instead of jumping. The Motion section of the inspector picks one of four springs, each set by how long it takes and how much it overshoots: Quick, Gentle, Bouncy and Slow. Its curve is drawn under the choice, and Play shows the part's motion in the preview. Panels choose how they come in: rise, pop or tilt. For movement over time, Animate on a timeline opens the keyframe editor with the part selected, so a panel can slide in during a film.","u":"/blog/design-panels-for-headsets-phones-and-glasses#motion-with-springs"},{"k":"blog","t":"Design: one editor for the panels in your 3D scene","h":"A theme for every part","x":"With nothing selected, the inspector shows the theme: the colours, spacing and motion every part falls back to. Change the accent colour there and every slider, toggle and button that uses it changes with it. A part's colour can follow a theme colour, or have its own.","u":"/blog/design-panels-for-headsets-phones-and-glasses#a-theme-for-every-part"},{"k":"blog","t":"Design: one editor for the panels in your 3D scene","h":"Checks for the device","x":"A panel that reads well on a monitor can be too small, too far or too high in a headset. Design's checks measure the panel where it stands, from where a person stands when the session starts: - In a headset: whether a finger can reach it, whether its smallest letters are large enough for their distance, whether its controls are big enough for a fingertip or for pointing, whether it fits in view, a","s":"A panel that reads well on a monitor can be too small, too far or too high in a headset. Design's checks measure the panel where it stands, from where a person stands when the session starts: - In a headset: whether a finger can reach it, whether its smallest letters are large enough for their distance, whether its controls are big enough for a fingertip or for pointing, whether it fits in view, and whether it sits between 10° above and 40° below eye level. - On a phone in AR: the smallest letters and controls in points on the screen. Click a check to select the part it is about, or ask the agent to fix it. The request already holds the numbers, such as \"make the smallest text 1.4 cm tall\". The same panel checked for a phone in AR.","u":"/blog/design-panels-for-headsets-phones-and-glasses#checks-for-the-device"},{"k":"blog","t":"Design: one editor for the panels in your 3D scene","h":"One panel, several platforms","x":"A panel can be written for other platforms from the inspector: - SwiftUI for Apple Vision Pro: stacks, sliders, toggles, pickers and buttons on a glass background, one point to a millimetre. - Unity UI Toolkit for Quest, Android XR and other headsets built with Unity: a UXML layout and a style sheet for a world-space panel. - Display glasses for Meta Ray-Ban Display: a screen built from Meta's Wea","s":"A panel can be written for other platforms from the inspector: - SwiftUI for Apple Vision Pro: stacks, sliders, toggles, pickers and buttons on a glass background, one point to a millimetre. - Unity UI Toolkit for Quest, Android XR and other headsets built with Unity: a UXML layout and a style sheet for a world-space panel. - Display glasses for Meta Ray-Ban Display: a screen built from Meta's Wearables UI toolkit. Glasses have no sliders or switches, so a slider steps with two buttons and a toggle says its state. Values your code binds, such as a slider's value={volume} , become state of the same name. What a button does is copied as a comment, for you or the agent to connect. We compiled the SwiftUI against the visionOS SDK and the glasses screen against Meta's toolkit before shipping it. The starter panel written as SwiftUI for Apple Vision Pro.","u":"/blog/design-panels-for-headsets-phones-and-glasses#one-panel-several-platforms"},{"k":"blog","t":"Design: one editor for the panels in your 3D scene","h":"In any web project","x":"Design started with the Hands & Spatial UI template. Other web projects can now add its UI kit: open Design, choose Add the UI kit , and Graspable puts it in a folder of its own and asks the agent for a first panel. Tabletop AR on a phone, React XR scenes, product viewers and games get the same editor.","u":"/blog/design-panels-for-headsets-phones-and-glasses#in-any-web-project"},{"k":"blog","t":"Design: one editor for the panels in your 3D scene","h":"What it does not do yet","x":"Design edits the parts of the UI kit. A panel built from plain meshes and shapes is still changed in code or by the agent. The checks are rules taken from comfort guidelines and our own measurements of text in WebXR, applied to the sizes in your scene; they are not a test on a device. And the code written for other platforms is a starting point: it places the parts and their values, and leaves the","s":"Design edits the parts of the UI kit. A panel built from plain meshes and shapes is still changed in code or by the agent. The checks are rules taken from comfort guidelines and our own measurements of text in WebXR, applied to the sizes in your scene; they are not a test on a device. And the code written for other platforms is a starting point: it places the parts and their values, and leaves the wiring to you or the agent. Design is in Graspable 0.21. Open a project from the Hands & Spatial UI template, or add the kit to your own, and choose Design in the header.","u":"/blog/design-panels-for-headsets-phones-and-glasses#what-it-does-not-do-yet"},{"k":"blog","t":"A million Gaussian splats on WebGPU, now in the headset","x":"The WebGPU Splats template draws splats with Radix, which sorts every splat on the graphics card each frame. Since 0.20.1 a headset browser that offers WebGPU for immersive sessions draws both eyes on WebGPU too. A new Graspable project from the WebGPU Splats template opens on a spiral galaxy of 600 000 Gaussian splats, made by code that runs on the graphics card. The renderer behind it is Radix, ","s":"The WebGPU Splats template draws splats with Radix, which sorts every splat on the graphics card each frame. Since 0.20.1 a headset browser that offers WebGPU for immersive sessions draws both eyes on WebGPU too. A new Graspable project from the WebGPU Splats template opens on a spiral galaxy of 600 000 Gaussian splats, made by code that runs on the graphics card. The renderer behind it is Radix, the WebGPU splat renderer from the team that makes Spark. Since Graspable 0.20.1, the same scene also draws on WebGPU inside a headset whose browser offers it.","u":"/blog/webgpu-splats-in-the-headset"},{"k":"blog","t":"A million Gaussian splats on WebGPU, now in the headset","h":"Sorting on the graphics card","x":"Gaussian splats are see-through, so they have to be drawn back to front, and the order changes whenever the viewer moves. Sorting is the expensive part. Spark, which draws our world templates, sorts on the processor in a worker and sends the order to the graphics card. On WebGPU, Radix keeps the whole frame on the graphics card. A projection pass places every splat relative to the viewer and write","s":"Gaussian splats are see-through, so they have to be drawn back to front, and the order changes whenever the viewer moves. Sorting is the expensive part. Spark, which draws our world templates, sorts on the processor in a worker and sends the order to the graphics card. On WebGPU, Radix keeps the whole frame on the graphics card. A projection pass places every splat relative to the viewer and writes a sort key. Splats outside the view are dropped, a radix sort orders the rest, and one draw call puts every splat of every object on screen. Radix's own explanation says nothing is read back to the processor. Radix's authors measured it against Spark across five scenes of 0.5 to 10 million splats. On WebGPU, Radix was 1.37 times as fast as Spark on WebGL2 on average, and 1.8 times as fast at 10 million. On WebGL2, Spark was slightly faster. Those are their numbers, on their hardware. Ours, on an Apple M1 Max in Chrome at 1600 × 1000, with frames paced by the graphics card: - 500 000 splats: 4.9 ms a frame - 1 000 000 splats: 10 to 12 ms - 2 000 000 splats: 17.6 ms Above about three million the frame times start to jump around, so that is the practical ceiling on that computer. A recent computer shows one to two million splats smoothly.","u":"/blog/webgpu-splats-in-the-headset#sorting-on-the-graphics-card"},{"k":"blog","t":"A million Gaussian splats on WebGPU, now in the headset","h":"Splats made by code","x":"In Radix a generator is a small shader that makes splat number i while the frame is drawn. A splat file is one generator; the galaxy is another. Changing a value such as the galaxy's rotation compiles nothing, so a scene can animate millions of splats every frame. Julia, 128 BPM shows how far that goes. It is a quaternion Julia fractal made of a million splats on WebGPU. The graphics card traces t","s":"In Radix a generator is a small shader that makes splat number i while the frame is drawn. A splat file is one generator; the galaxy is another. Changing a value such as the galaxy's rotation compiles nothing, so a scene can animate millions of splats every frame. Julia, 128 BPM shows how far that goes. It is a quaternion Julia fractal made of a million splats on WebGPU. The graphics card traces the shape into a table, traces a slice of it again each frame as the music changes it, and lights every sample every frame. The project was rebuilt on Radix in Graspable and is published in Made in Graspable. The template also reads captured places: .spz and .ply files, including compressed .ply , drawn in the same sort as the splats made by code.","u":"/blog/webgpu-splats-in-the-headset#splats-made-by-code"},{"k":"blog","t":"A million Gaussian splats on WebGPU, now in the headset","h":"Both eyes on WebGPU","x":"A headset session needs the browser to hand the page a WebGPU surface for each eye. That is the WebXR/WebGPU Binding, still an editor's draft. Meta Browser 146 added experimental support for WebGPU in WebXR, and three.js 0.186, which the template uses, can render each eye into it. In 0.20.1 the template asks for a WebGPU session wherever the browser offers one. Radix then projects and sorts once p","s":"A headset session needs the browser to hand the page a WebGPU surface for each eye. That is the WebXR/WebGPU Binding, still an editor's draft. Meta Browser 146 added experimental support for WebGPU in WebXR, and three.js 0.186, which the template uses, can render each eye into it. In 0.20.1 the template asks for a WebGPU session wherever the browser offers one. Radix then projects and sorts once per frame from the head, and only the draw is done twice, once per eye. A browser without the binding gets Radix's WebGL2 path, where the projection runs on the graphics card and the sort runs in a worker. The choice is made once, before the first frame, because a canvas cannot change its type later. To use it on a Quest, run Meta Browser 146 or later, turn on its WebGPU and WebXR experiments in chrome://flags , and restart the browser. The difference should be largest on the headset. On the WebGL2 path the headset's processor does the sort, and it is much slower than a desktop's. On WebGPU it does not take part. Open Julia, 128 BPM in the headset: under the fractal it shows the path in use, the splat count, the frame rate and the frame time. Add ?gpu=webgpu or ?gpu=webgpu-gl to the address to compare the two paths, and ?count= to change the number of splats.","u":"/blog/webgpu-splats-in-the-headset#both-eyes-on-webgpu"},{"k":"blog","t":"A million Gaussian splats on WebGPU, now in the headset","h":"Starting a project","x":"Create a project from WebGPU Splats in Graspable and describe what you want: replace the galaxy with your own shape, load a captured room from public/splats/ , or mix both with ordinary three.js meshes. The template comes with a guide the agent reads before it writes splat code. The preview shows the scene on your screen and in the emulated headset, and a published app opens in a headset browser a","s":"Create a project from WebGPU Splats in Graspable and describe what you want: replace the galaxy with your own shape, load a captured room from public/splats/ , or mix both with ordinary three.js meshes. The template comes with a guide the agent reads before it writes splat code. The preview shows the scene on your screen and in the emulated headset, and a published app opens in a headset browser at its own address. Radix is new (version 0.0.1). It does not yet read .splat , .ksplat or .sog files, has no level-of-detail streaming and cannot pick a splat with a pointer. For large captured worlds with streaming, the world templates on Spark are still the right choice.","u":"/blog/webgpu-splats-in-the-headset#starting-a-project"},{"k":"blog","t":"A million Gaussian splats on WebGPU, now in the headset","h":"Sources","x":"- Radix and why Radix, with its benchmark against Spark - WebXR/WebGPU Binding Module, editor's draft - Meta Browser 146 release notes - Graspable changelog: 0.20.1","u":"/blog/webgpu-splats-in-the-headset#sources"},{"k":"blog","t":"Graspable 0.19.8: gestures, the real room, and AI inside your apps","x":"Graspable 0.19.8 gives web projects hand and head gestures and physics that meets your real furniture, and lets a published app ask an AI model about what it sees, with no key anywhere in the app. Graspable 0.19.8 is about apps that notice you. A new React XR project can tell a thumbs-up from a pinch, a ball you throw in AR lands on your real table, and once you publish, the app can ask an AI mode","s":"Graspable 0.19.8 gives web projects hand and head gestures and physics that meets your real furniture, and lets a published app ask an AI model about what it sees, with no key anywhere in the app. Graspable 0.19.8 is about apps that notice you. A new React XR project can tell a thumbs-up from a pinch, a ball you throw in AR lands on your real table, and once you publish, the app can ask an AI model what is in front of it and answer out loud. The parts are small and sit in the project as ordinary files, so the agent uses them when you ask for something that needs them, and you can read every line.","u":"/blog/gestures-real-room-and-ai-in-your-apps"},{"k":"blog","t":"Graspable 0.19.8: gestures, the real room, and AI inside your apps","h":"Hands that mean something","x":"Until now, hands in our web templates could point, grab and press. That covers a lot, but people also make shapes with their hands and expect something to happen. The React XR and XR Game templates now recognise six of them: pinch, point, fist, thumbs-up, open palm and spread. They are worked out from the joints the headset already reports, every frame, with a short hold before a gesture counts an","s":"Until now, hands in our web templates could point, grab and press. That covers a lot, but people also make shapes with their hands and expect something to happen. The React XR and XR Game templates now recognise six of them: pinch, point, fist, thumbs-up, open palm and spread. They are worked out from the joints the headset already reports, every frame, with a short hold before a gesture counts and a lower threshold before it ends, so a hand halfway between two shapes does not flicker. Heads count too. A nod and a shake are recognised from the headset's own motion, while looking around and slow turns are ignored. That gives you a yes and a no without a single button. In a new React XR project, a thumbs-up adds a ball to the bench. In XR Game, a thumbs-up after a round starts the next one. In your own code it is one hook: Or skip the code and ask: \"open the menu when I show an open palm, close it with a fist.\"","u":"/blog/gestures-real-room-and-ai-in-your-apps#hands-that-mean-something"},{"k":"blog","t":"Graspable 0.19.8: gestures, the real room, and AI inside your apps","h":"Physics that meets your room","x":"In AR, a Quest 3 or an Android XR headset knows where your floor, table and walls are. The React XR template now turns those surfaces and the room's scanned shape into solid ground for the physics, and updates them when the headset refines its picture of the room. Drop a ball over your desk and it lands on the desk. Where the headset's browser offers depth, the scene can also hide behind real thin","s":"In AR, a Quest 3 or an Android XR headset knows where your floor, table and walls are. The React XR template now turns those surfaces and the room's scanned shape into solid ground for the physics, and updates them when the headset refines its picture of the room. Drop a ball over your desk and it lands on the desk. Where the headset's browser offers depth, the scene can also hide behind real things: your hand passes in front of a virtual cube instead of being drawn over by it. It is one line when the store is created, createXRStore(withOcclusion({})) , and devices without depth simply skip it.","u":"/blog/gestures-real-room-and-ai-in-your-apps#physics-that-meets-your-room"},{"k":"blog","t":"Graspable 0.19.8: gestures, the real room, and AI inside your apps","h":"AI inside a published app","x":"This is the one I have wanted for a while. An app made in Graspable can now ask an AI model a question, with a picture of what it is showing, and hear the answer. The usual way to do this is to put an API key in the app, which means anyone who opens it can read the key and spend on it. Graspable apps hold no key. The app sends its request to its own address, and Graspable asks the model on your be","s":"This is the one I have wanted for a while. An app made in Graspable can now ask an AI model a question, with a picture of what it is showing, and hear the answer. The usual way to do this is to put an API key in the app, which means anyone who opens it can read the key and spend on it. Graspable apps hold no key. The app sends its request to its own address, and Graspable asks the model on your behalf, with the model and the fixed instruction you chose for that app. Visitors sign in to nothing. You pay from your Credits, at what the model costs, under a daily limit you set, and the app stops asking when the limit is reached. Every web template has the helper: A chemistry set that explains the reaction you just started, a museum piece that answers questions about itself, a puzzle that gives hints when you nod at it: these are a request away now. Tell the agent what the app should ask and when, and it uses the helper. Then switch AI on for the app on graspable.dev, under Account → Published apps , and pick the model. It comes with Builder and higher, and works in the preview on your computer as soon as the project is published. What we keep is counts: how many requests an app made and what they cost. Never the questions, pictures or answers.","u":"/blog/gestures-real-room-and-ai-in-your-apps#ai-inside-a-published-app"},{"k":"blog","t":"Graspable 0.19.8: gestures, the real room, and AI inside your apps","h":"Where to start","x":"Create a React XR project and try the thumbs-up, or open one you already have and ask the agent to add gestures or the real room. The details are in the docs: gestures and the real room and AI in your app. The full list of changes is in the changelog. The app updates itself, and new installs are on the download page.","u":"/blog/gestures-real-room-and-ai-in-your-apps#where-to-start"},{"k":"blog","t":"Graspable 0.19.7: model in Blender and build the scene at the same time","x":"Graspable 0.19.7 lets the agent split a request between helpers that work side by side, so a model can take shape in Blender while its interactions are written. Tidal Observatory is now a room you walk into, built that way. In the 0.19.6 post, Tidal Observatory was a brass tide instrument on a table. Today you can walk into it: a round observatory with a dome that opens onto the moon, a refracting","s":"Graspable 0.19.7 lets the agent split a request between helpers that work side by side, so a model can take shape in Blender while its interactions are written. Tidal Observatory is now a room you walk into, built that way. In the 0.19.6 post, Tidal Observatory was a brass tide instrument on a table. Today you can walk into it: a round observatory with a dome that opens onto the moon, a refracting telescope you aim by its handle, and an armillary sphere whose rings you spin with your hands. Everything in the room was modelled in Blender, on the same computer, during the same request that wrote the code to use it. Graspable 0.19.7 makes that possible. The agent can hand parts of a request to helpers, and helpers started together now work at the same time. For a scene like this one, the Blender model and the interactions are built in parallel, and you get the room in roughly the time the model takes. The full request, recorded: Graspable on the left, Blender on the right. Slow stretches are sped up where marked. At the end, the observatory in an emulated Quest 3, used with controllers and with bare hands.","u":"/blog/walk-in-observatory-blender-and-graspable"},{"k":"blog","t":"Graspable 0.19.7: model in Blender and build the scene at the same time","h":"Step inside Tidal Observatory","x":"Open Tidal Observatory in a browser or on a headset. In VR the room stands around you, with the original instrument still on its plinth in the middle. Pull the brass lever on your left and the dome's shutters slide apart onto a night sky. In Blender they are one animation, four seconds long, and the scene plays it straight from the exported model. The open dome, seen from the middle of the room. T","s":"Open Tidal Observatory in a browser or on a headset. In VR the room stands around you, with the original instrument still on its plinth in the middle. Pull the brass lever on your left and the dome's shutters slide apart onto a night sky. In Blender they are one animation, four seconds long, and the scene plays it straight from the exported model. The open dome, seen from the middle of the room. Take the telescope by its wooden grip and swing it. It turns and tilts on its own mount, a little heavy, and stays where you let go. A round view above the mount shows what the tube is pointing at, and when the moon is in it, a ring around the view lights amber. The moon in the telescope's view. The armillary sphere works with controllers or bare hands. Pinch a ring, turn it and let go: it keeps spinning and slowly comes to rest. On a flat screen, Open dome and Spin sphere sit in the toolbar. In AR the walls disappear, and the instrument and its three machines stand in your own room. The whole scene is about 199,000 triangles with 1024 px photographed textures. Continue from it and you get both Blender files and the Python scripts that build them, so you can change a dimension and rebuild, or open the .blend and edit it by hand.","u":"/blog/walk-in-observatory-blender-and-graspable#step-inside-tidal-observatory"},{"k":"blog","t":"Graspable 0.19.7: model in Blender and build the scene at the same time","h":"Blender and code, side by side","x":"Models are usually the slow part of a 3D scene. Until this release, the code that brings them to life waited until the model was finished. In 0.19.7 the two run in parallel. What makes that work is agreeing on names first. Before any modelling started, the agent wrote a short file, source/room-contract.md , listing every moving part with its name, pivot, position in metres and animation. One helpe","s":"Models are usually the slow part of a 3D scene. Until this release, the code that brings them to life waited until the model was finished. In 0.19.7 the two run in parallel. What makes that work is agreeing on names first. Before any modelling started, the agent wrote a short file, source/room-contract.md , listing every moving part with its name, pivot, position in metres and animation. One helper modelled the room in Blender to that list. The other wrote the grabbing, spinning, the lever and the telescope's view against the same names, using simple boxes and cylinders in their place. When the model arrived, it dropped in where the boxes had been. The room in Blender, seen through the dome. Blender reloaded the file each time it was saved. For this room the interactions were ready in under three minutes. Blender took about fourteen, including several rebuilds after reading its own error log, Blender's asset check, review renders from four sides, and a check that every exported part sat where the list said. The whole request, from typing it to a tested scene, took 22 minutes. Each helper's Blender commands and file changes asked for approval like the agent's own, and its steps appeared in the same run.","u":"/blog/walk-in-observatory-blender-and-graspable#blender-and-code-side-by-side"},{"k":"blog","t":"Graspable 0.19.7: model in Blender and build the scene at the same time","h":"Tested with controllers and hands","x":"Every Graspable project carries an emulated Meta Quest 3, and the scene was tried in it before the request ended: pointing at the lever to open the dome, gripping the telescope's handle with a controller, and switching to hands to pinch and turn a ring. Because the room changed the scene, the headset recordings saved with the project no longer matched, and Graspable asked before keeping the new sc","s":"Every Graspable project carries an emulated Meta Quest 3, and the scene was tried in it before the request ended: pointing at the lever to open the dome, gripping the telescope's handle with a controller, and switching to hands to pinch and turn a ring. Because the room changed the scene, the headset recordings saved with the project no longer matched, and Graspable asked before keeping the new scene as the reference for later checks. The emulator tells you that the scene loads, runs and answers controllers and hands. How smoothly it runs on a real Quest 3 is something to check on the headset.","u":"/blog/walk-in-observatory-blender-and-graspable#tested-with-controllers-and-hands"},{"k":"blog","t":"Graspable 0.19.7: model in Blender and build the scene at the same time","h":"Try it on your own scene","x":"Ask the Graspable agent for helpers in your request and say what each one owns: one for the Blender model, one for the scene, each with its own files. Name the moving parts you need, or ask the agent to write them down before the helpers start. A request can use up to three helpers, and helpers you ask for in the same step work at the same time. The Blender skills from the skill library give the m","s":"Ask the Graspable agent for helpers in your request and say what each one owns: one for the Blender model, one for the scene, each with its own files. Name the moving parts you need, or ask the agent to write them down before the helpers start. A request can use up to three helpers, and helpers you ask for in the same step work at the same time. The Blender skills from the skill library give the model helper what it needs to know about Blender 5, glTF export and asset checks. Graspable uses the copy of Blender already installed on your computer. The full list of changes is in the changelog. The app updates itself, and new installs are on the download page.","u":"/blog/walk-in-observatory-blender-and-graspable#try-it-on-your-own-scene"},{"k":"blog","t":"Graspable 0.19.6: a Blender model in a WebXR scene, and a faster USD viewer","x":"How the Graspable agent used a local copy of Blender to model, texture, animate and export Tidal Observatory, how the result is checked in an emulated headset, and why USD scenes now open without conversion. Graspable 0.19.6 is out. It opens USD scenes directly in the app, and it is the release we use to show something that already works: the agent can drive the copy of Blender on your computer, k","s":"How the Graspable agent used a local copy of Blender to model, texture, animate and export Tidal Observatory, how the result is checked in an emulated headset, and why USD scenes now open without conversion. Graspable 0.19.6 is out. It opens USD scenes directly in the app, and it is the release we use to show something that already works: the agent can drive the copy of Blender on your computer, keep the .blend file editable, and put the exported model into a WebXR scene that it then checks in an emulated headset. Tidal Observatory is the example. Open it in a browser or a headset, or continue from it to get the Blender file and the script that builds it. A ten-minute agent run, recorded: Graspable on the left, Blender on the right reloading each .blend the agent saves. Sped up where marked.","u":"/blog/blender-to-webxr-tidal-observatory"},{"k":"blog","t":"Graspable 0.19.6: a Blender model in a WebXR scene, and a faster USD viewer","h":"One request, made in Blender","x":"Tidal Observatory started as one request to the Graspable agent in a new React XR project. The request named the Blender program on the computer, asked for a tabletop tide instrument in brass, ivory and midnight blue, and set the limits: an editable .blend , a script that rebuilds it, a .glb with named parts and baked animation, under 150,000 triangles, and a scene that still works in VR and AR. T","s":"Tidal Observatory started as one request to the Graspable agent in a new React XR project. The request named the Blender program on the computer, asked for a tabletop tide instrument in brass, ivory and midnight blue, and set the limits: an editable .blend , a script that rebuilds it, a .glb with named parts and baked animation, under 150,000 triangles, and a scene that still works in VR and AR. The agent read six Blender skills first. A skill is a short set of instructions for one kind of task; these cover Blender's Python interface as it is in version 5, materials that survive glTF export, baked animation, export settings, asset checks and lighting. They are in the skill library and you add them to a project like any other skill. Then it wrote source/build observatory.py and ran Blender without a window: The script models every part from scratch, assigns Principled materials, keys the eight aperture leaves and the orbiting moon, saves source/tidal-observatory.blend and exports public/assets/tidal-observatory.glb . Each Blender command was a step the person approves, like any other command. After each build the agent ran the asset check from the asset QA skill inside Blender: open edges, flipped normals, scale, triangle and material counts. It also rendered review sheets from four sides, in materials and in wireframe, and looked at them before going on. One of the review sheets the agent rendered in Blender and inspected before exporting. The whole first version took one run of 39 minutes: 92 steps, ten of them commands, six of them checks of the scene in the app. Two later changes made the scene you can open today. A second pass replaced the procedural brass with photographed surfaces from Poly Haven (CC0), put photographed dark wood on the base, added a clear lacquer to the blue body, and lit the scene in the browser with a photographed studio. A third request added a telescope; that one is in the film above.","u":"/blog/blender-to-webxr-tidal-observatory#one-request-made-in-blender"},{"k":"blog","t":"Graspable 0.19.6: a Blender model in a WebXR scene, and a faster USD viewer","h":"A change, recorded","x":"The film at the top is a second request on the finished scene: add a small brass telescope on a three-legged stand, aimed at the moon, built in Blender with the instrument's own materials, then checked in the emulated headset. On the left is Graspable; on the right, Blender with the project's .blend open, reloading the file each time the agent saves it. The agent changed build observatory.py and r","s":"The film at the top is a second request on the finished scene: add a small brass telescope on a three-legged stand, aimed at the moon, built in Blender with the instrument's own materials, then checked in the emulated headset. On the left is Graspable; on the right, Blender with the project's .blend open, reloading the file each time the agent saves it. The agent changed build observatory.py and ran Blender. The first build failed, so it read Blender's log, found the cause in its own script and built again. Blender's asset check passed, and the telescope added 1,732 triangles and no new materials. It then tested the scene in the emulated Quest 3: aimed a controller at the instrument's Close aperture button, pressed it, pressed it again and looked at the telescope from above to see that its legs stand on the plinth. The run took ten minutes. The scene changed, so the headset recordings saved with the project no longer matched. Graspable flagged that difference and asked whether to keep the new scene as the reference for later checks, rather than accepting it by itself.","u":"/blog/blender-to-webxr-tidal-observatory#a-change-recorded"},{"k":"blog","t":"Graspable 0.19.6: a Blender model in a WebXR scene, and a faster USD viewer","h":"The Blender file stays yours","x":"Nothing is locked inside a model file. source/build observatory.py is ordinary Blender Python, so you can read why each part has its size, change a number, and ask the agent to rebuild. You can also open source/tidal-observatory.blend in Blender and edit it by hand. The README the agent wrote says which is the source: a hand edit in the .blend is overwritten by the next rebuild, so keep it, or mov","s":"Nothing is locked inside a model file. source/build observatory.py is ordinary Blender Python, so you can read why each part has its size, change a number, and ask the agent to rebuild. You can also open source/tidal-observatory.blend in Blender and edit it by hand. The README the agent wrote says which is the source: a hand edit in the .blend is overwritten by the next rebuild, so keep it, or move it into the script. With the telescope, the exported model has about 90,500 triangles and 9 materials, with two animations: the aperture (1.6 seconds) and the moon's orbit (24 seconds). The scene plays them from the .glb , so what moves in the browser is what was keyed in Blender.","u":"/blog/blender-to-webxr-tidal-observatory#the-blender-file-stays-yours"},{"k":"blog","t":"Graspable 0.19.6: a Blender model in a WebXR scene, and a faster USD viewer","h":"Checked in an emulated headset","x":"A Blender render does not show how a scene behaves in a browser or a headset. So after export the agent checks the real page. Every Graspable project carries an emulated Meta Quest 3: the agent loads the built scene in it, points a controller at a button, presses it and reads what changed in the scene. The run ends only when the build passes and the page loads without errors. The scene in the emul","s":"A Blender render does not show how a scene behaves in a browser or a headset. So after export the agent checks the real page. Every Graspable project carries an emulated Meta Quest 3: the agent loads the built scene in it, points a controller at a button, presses it and reads what changed in the scene. The run ends only when the build passes and the page loads without errors. The scene in the emulated Quest 3 in Graspable's preview, with the device controls beside it. The emulator answers whether the scene runs, loads its model and responds to a controller. It does not tell you how the scene feels on your head or how fast it runs on the headset's chip. For that, open the published address on the headset itself.","u":"/blog/blender-to-webxr-tidal-observatory#checked-in-an-emulated-headset"},{"k":"blog","t":"Graspable 0.19.6: a Blender model in a WebXR scene, and a faster USD viewer","h":"USD scenes open directly","x":"0.19.6 also changes how the app shows OpenUSD scenes. Until now, opening a .usd , .usda , .usdc or .usdz file in the Assets tab converted it to a glTF model first, with the OpenUSD runtime that comes with Graspable, and showed nothing where that runtime was missing. The app now reads the scene itself with @huggingface/usd-web-viewer, Hugging Face's OpenUSD viewer for the web. It composes the USD l","s":"0.19.6 also changes how the app shows OpenUSD scenes. Until now, opening a .usd , .usda , .usdc or .usdz file in the Assets tab converted it to a glTF model first, with the OpenUSD runtime that comes with Graspable, and showed nothing where that runtime was missing. The app now reads the scene itself with @huggingface/usd-web-viewer, Hugging Face's OpenUSD viewer for the web. It composes the USD layers in WebAssembly with the openusd Rust library and draws them with three.js, in a worker so the app stays responsive. Graspable hands it the scene and every file the scene refers to, so references, payloads and textures resolve as they do on disk. A SimReady fire extinguisher or pallet from NVIDIA's library opens in about half a second. The viewer's source is in the huggingface.js repository, under the MIT licence. The viewer does not play animation or skeletons yet. When a scene has a time range, the app says so and offers Play its animation , which opens the converted preview as before. Editing USD, layers and export are unchanged; they still use the official OpenUSD libraries.","u":"/blog/blender-to-webxr-tidal-observatory#usd-scenes-open-directly"},{"k":"blog","t":"Graspable 0.19.6: a Blender model in a WebXR scene, and a faster USD viewer","h":"Also in 0.19.6","x":"- Copy buttons for installer checksums and crypto payment details confirm only after copying worked, and show the full text to copy by hand when the browser blocks the clipboard. The full list is in the changelog. The app updates itself; new installs are on the download page.","u":"/blog/blender-to-webxr-tidal-observatory#also-in-0-19-6"},{"k":"blog","t":"Build an editable WebXR exhibit with your coding agent","x":"An assembly study in one React Three Fiber component: select parts, separate them and inspect them in 3D or WebXR. Its source is below, and your coding agent can apply it through Graspable’s MCP tools. Your coding agent can edit a WebXR project directly through Graspable. It supplies the code; Graspable creates the project, builds it, checks that it loads in a browser and opens a preview. You do n","s":"An assembly study in one React Three Fiber component: select parts, separate them and inspect them in 3D or WebXR. Its source is below, and your coding agent can apply it through Graspable’s MCP tools. Your coding agent can edit a WebXR project directly through Graspable. It supplies the code; Graspable creates the project, builds it, checks that it loads in a browser and opens a preview. You do not need to hand the request to a second AI agent. Here is a small example you can reproduce: a fictional beacon on an inspection table. Separate its three parts, select one to read about it, rotate the parts, then put everything back. The source is one component in the free React XR template. The picture above is a real desktop-browser capture. The beacon is made from editable geometry; it is not an engineering model. The whole component is at the end of this article. It uses only what the React XR template already installs: no models, textures or generation services to buy.","u":"/blog/build-webxr-with-an-external-agent"},{"k":"blog","t":"Build an editable WebXR exhibit with your coding agent","h":"Start with a connected coding agent","x":"Download Graspable, then install its agent connection from Settings → Plugins and connectors → Use Graspable from other tools . Ask your coding agent to follow the MCP setup guide. Claude Code, Codex and other clients that can start an MCP server can use this connection. Their setup steps differ. Once connected, give the agent this request: Create a new free React XR project called Beacon inspecti","s":"Download Graspable, then install its agent connection from Settings → Plugins and connectors → Use Graspable from other tools . Ask your coding agent to follow the MCP setup guide. Claude Code, Codex and other clients that can start an MCP server can use this connection. Their setup steps differ. Once connected, give the agent this request: Create a new free React XR project called Beacon inspection. Read https://graspable.dev/blog/build-webxr-with-an-external-agent.md and the project's current components/Workbench.jsx. Replace that file with the component under \"The whole component\". Keep the template's application, XR setup and other files unchanged. Use Graspable's MCP tools to check the project and open its preview. Do not start a delegated AI run, purchase anything or publish the project. Report the checks that passed and any interactions you could not test. If you are applying the file yourself, create a new React XR project, open Code , and replace components/Workbench.jsx with the source at the end of this article. Use a new project so the example does not overwrite an exhibit you have already built. Creating this free project, editing its files and checking it locally do not spend Graspable Credits or require another AI sign-in. Your coding agent's own subscription or API usage terms still apply. Publishing and paid services are separate choices.","u":"/blog/build-webxr-with-an-external-agent#start-with-a-connected-coding-agent"},{"k":"blog","t":"Build an editable WebXR exhibit with your coding agent","h":"How the component works","x":"The scene is three parts described as data. Each entry is a name, the sentence shown when it is selected, and a colour: Adding a part to the data adds it to the table, the selection buttons and the sign above the table. Each part moves towards where it should be on every frame, so separating, assembling and rotating are smooth without an animation library. MathUtils.damp eases the position, and pe","s":"The scene is three parts described as data. Each entry is a name, the sentence shown when it is selected, and a colour: Adding a part to the data adds it to the table, the selection buttons and the sign above the table. Each part moves towards where it should be on every frame, so separating, assembling and rotating are smooth without an animation library. MathUtils.damp eases the position, and people who ask their system for reduced motion get the end position at once: The same state drives two sets of controls. On a flat screen an HTML panel holds buttons that work with a mouse, Tab and Enter. In a headset that panel is gone, and buttons on the table do the same things: Label and SceneButton are components the React XR template already has, in the same components folder: text drawn for the headset, and a button that can be pressed with a controller, a hand or a mouse.","u":"/blog/build-webxr-with-an-external-agent#how-the-component-works"},{"k":"blog","t":"Build an editable WebXR exhibit with your coding agent","h":"What the agent does","x":"The reproducible MCP sequence is: 1. list templates , then create project with react-xr-developer . 2. Wait until list projects reports that installation is done. 3. read file for components/Workbench.jsx , then write file with the replacement. Read it back to check that the intended source was saved. 4. check project . Correct build or browser errors before continuing. 5. start preview , then ins","s":"The reproducible MCP sequence is: 1. list templates , then create project with react-xr-developer . 2. Wait until list projects reports that installation is done. 3. read file for components/Workbench.jsx , then write file with the replacement. Read it back to check that the intended source was saved. 4. check project . Correct build or browser errors before continuing. 5. start preview , then inspect the running scene at the returned local address. For this example we authored the component, refined it while viewing the scene, and applied the final file to a fresh project through Graspable 0.18.1's MCP tools. The file readback, build, browser check and preview passed on macOS. This was a direct-edit test with no delegated model run. It was not a test of every named client's connection or a claim that one prompt always produces this result. A successful check proves less than a working exhibit. Our first build passed while its desktop controls were below the viewport. Looking at the actual preview caught that mistake. After correcting it, we tested the controls themselves.","u":"/blog/build-webxr-with-an-external-agent#what-the-agent-does"},{"k":"blog","t":"Build an editable WebXR exhibit with your coding agent","h":"Try the interactions","x":"Select Separate parts . The optics, power core and radiator move apart. Choose a part in the control panel, or click its geometry. The sign and selection ring should agree with your choice. Choose Rotate parts to inspect the shapes, then Face front to return them to their starting orientation. Reset assembles the parts, stops rotation and selects the optics. The Inspect the beacon heading hides or","s":"Select Separate parts . The optics, power core and radiator move apart. Choose a part in the control panel, or click its geometry. The sign and selection ring should agree with your choice. Choose Rotate parts to inspect the shapes, then Face front to return them to their starting orientation. Reset assembles the parts, stops rotation and selects the optics. The Inspect the beacon heading hides or shows the desktop controls. Those controls also work with Tab and Enter. In an immersive session, the HTML panel disappears. Buttons on the table provide separation, rotation, reset and Leave XR . We exercised selection, separation and exit using an emulated controller in IWER, as well as desktop mouse and keyboard controls. A narrow browser layout was also checked at 390 pixels wide. This example has not had a physical-headset or phone performance test. Emulation does not establish hand tracking, readable text in a headset or sustained frame rate. It does not detect a real table on iPhone: this is a WebXR inspection exhibit, not a markerless phone-AR placement example. For that distinction, see AR on phones and headsets.","u":"/blog/build-webxr-with-an-external-agent#try-the-interactions"},{"k":"blog","t":"Build an editable WebXR exhibit with your coding agent","h":"Make a change that you can check","x":"Try a bounded follow-up: Add a fourth part called Protective housing. Give it its own selectable geometry and description, include it in separation and reset, and keep all four parts visible in a narrow browser window. Preserve the desktop keyboard controls and the in-world exit button. Check the build, then verify selection and reset in the preview. You retain the component and its state as ordin","s":"Try a bounded follow-up: Add a fourth part called Protective housing. Give it its own selectable geometry and description, include it in separation and reset, and keep all four parts visible in a narrow browser window. Preserve the desktop keyboard controls and the in-world exit button. Check the build, then verify selection and reset in the preview. You retain the component and its state as ordinary source. An agent can change the geometry, spacing, materials, descriptions and behavior; you can edit the same file yourself. For an actual product, replace the fictional parts with models you own and write descriptions based on that product. The useful handover is the editable project, the running preview and a clear record of what was tested. Use Graspable with your AI agent covers the connection, access limits and optional delegated workflow.","u":"/blog/build-webxr-with-an-external-agent#make-a-change-that-you-can-check"},{"k":"blog","t":"Build an editable WebXR exhibit with your coding agent","h":"The whole component","x":"This replaces components/Workbench.jsx in a new React XR project. The imports at the top are the template's own: React, React Three Fiber, drei, React XR and three are installed with it, and ./Label and ./SceneButton sit beside this file.","u":"/blog/build-webxr-with-an-external-agent#the-whole-component"},{"k":"blog","t":"Build an XR experience with NVIDIA OpenUSD assets","x":"Graspable 0.19 brings NVIDIA’s SimReady library to Pro. Import the editable source, inspect an object in VR, or turn a warehouse layout into a playable rehearsal. A pallet is an ordinary object until you need to build one. Then you need its dimensions, boards, nails, textures and materials. Multiply that work by a warehouse. Graspable 0.19 adds a searchable NVIDIA SimReady asset library to Pro and","s":"Graspable 0.19 brings NVIDIA’s SimReady library to Pro. Import the editable source, inspect an object in VR, or turn a warehouse layout into a playable rehearsal. A pallet is an ordinary object until you need to build one. Then you need its dimensions, boards, nails, textures and materials. Multiply that work by a warehouse. Graspable 0.19 adds a searchable NVIDIA SimReady asset library to Pro and Studio . Browse the OpenUSD collection on Hugging Face, inspect a package, and import its source into your project. Your agent can search and import from the same library. We used it to build two examples you can open now. <video controls playsInline preload=\"metadata\" poster=\"/site/motion/openusd-library-release.jpg\" <source src=\"/site/motion/openusd-library-release.mp4\" type=\"video/mp4\" / </video The film shows the actual examples running in the XR simulator, including controller input.","u":"/blog/nvidia-openusd-assets-in-graspable"},{"k":"blog","t":"Build an XR experience with NVIDIA OpenUSD assets","h":"Clear the aisle","x":"Open the warehouse rehearsal. Three pallets block the route to dispatch. In a supported WebXR headset, enter VR and use the controls in front of you to clear the aisle. Look around the room as each pallet reaches its bay. You can also play the animation from inside VR; its camera stays under your control. On a flat screen, select a pallet and its bay, or watch the edited camera rehearsal. The pall","s":"Open the warehouse rehearsal. Three pallets block the route to dispatch. In a supported WebXR headset, enter VR and use the controls in front of you to clear the aisle. Look around the room as each pallet reaches its bay. You can also play the animation from inside VR; its camera stays under your control. On a flat screen, select a pallet and its bay, or watch the edited camera rehearsal. The pallets and wall-mounted fire extinguisher come from NVIDIA’s collection. The room, crates, interaction and camera sequence were built around them in Graspable. The project keeps a 24-second take with three camera shots and editable pallet keyframes. Continue from this project to change the layout, timing or camera moves. For example, ask your agent: Keep the existing pallet models. Make the second pallet wait until the first reaches its bay. Add a closer camera shot of the last move. Keep the route markings visible. This is a small layout and animation example. It does not simulate forklift handling or validate workplace safety.","u":"/blog/nvidia-openusd-assets-in-graspable#clear-the-aisle"},{"k":"blog","t":"Build an XR experience with NVIDIA OpenUSD assets","h":"Look closely before you build around it","x":"Open the object study. Enter VR to see the extinguisher at its source size, hold it with a controller or pinch, and lift or turn it for a closer look. Reset returns it to the stand. On a flat screen, inspect the handle, gauge and surface label with the camera controls. Its displayed height comes from the source model’s bounds. Continue from the study to make a product explainer, a parts walkthroug","s":"Open the object study. Enter VR to see the extinguisher at its source size, hold it with a controller or pinch, and lift or turn it for a closer look. Reset returns it to the stand. On a flat screen, inspect the handle, gauge and surface label with the camera controls. Its displayed height comes from the source model’s bounds. Continue from the study to make a product explainer, a parts walkthrough or an inspection interface for another object. You can change the presentation while retaining the imported source package.","u":"/blog/nvidia-openusd-assets-in-graspable#look-closely-before-you-build-around-it"},{"k":"blog","t":"Build an XR experience with NVIDIA OpenUSD assets","h":"Find an asset without collecting the whole dataset","x":"Open a project, choose Assets → OpenUSD library , and search for an object. The catalog in this release contains 2,575 asset folders. Selecting one shows its preview, package size, file count and license before you import it. Choose Import and inspect . Graspable checks the package and its files, brings in the referenced sources, and includes the license and attribution. Imports are bounded: a pac","s":"Open a project, choose Assets → OpenUSD library , and search for an object. The catalog in this release contains 2,575 asset folders. Selecting one shows its preview, package size, file count and license before you import it. Choose Import and inspect . Graspable checks the package and its files, brings in the referenced sources, and includes the license and attribution. Imports are bounded: a package over 256 MiB, a file over 64 MiB, or an unsupported package layout is refused with a reason. You can still open its source page. The library is also available to the agent. Ask for an asset by what it is, then have the agent inspect what was imported before using it. Downloading an asset does not spend Graspable Credits; asking an AI provider to work with it follows your selected account’s billing.","u":"/blog/nvidia-openusd-assets-in-graspable#find-an-asset-without-collecting-the-whole-dataset"},{"k":"blog","t":"Build an XR experience with NVIDIA OpenUSD assets","h":"Source fidelity and web delivery are different jobs","x":"OpenUSD can describe more than a browser renderer supports. The original package stays in the project. A web conversion produces a separate model and a report describing unsupported or approximated features. Custom shaders, simulation settings and engine-specific behavior do not automatically become WebXR features. For these examples, we converted supported geometry and materials, resized textures","s":"OpenUSD can describe more than a browser renderer supports. The original package stays in the project. A web conversion produces a separate model and a report describing unsupported or approximated features. Custom shaders, simulation settings and engine-specific behavior do not automatically become WebXR features. For these examples, we converted supported geometry and materials, resized textures, and simplified the web copies. The pallet’s web model went from about 13 MB to 261 KB; the extinguisher went from about 12 MB to 1.6 MB. Those are file-size measurements, not frame-rate claims. The original USD files remain available in the editable projects. Both examples passed build and browser checks. We also checked their controls and rendering in an emulated headset. This release does not claim a new physical-headset performance benchmark for these scenes.","u":"/blog/nvidia-openusd-assets-in-graspable#source-fidelity-and-web-delivery-are-different-jobs"},{"k":"blog","t":"Build an XR experience with NVIDIA OpenUSD assets","h":"Open assets, a built-in workflow","x":"NVIDIA publishes these packages under CC BY 4.0. Their source is available independently of a Graspable plan. Pro and Studio include the integrated search, checked import and agent workflow. Attribution travels with the import; Graspable is not endorsed by NVIDIA or Hugging Face. Version 0.19 also brings the curated skills library into the desktop app, including project-specific adding and removin","s":"NVIDIA publishes these packages under CC BY 4.0. Their source is available independently of a Graspable plan. Pro and Studio include the integrated search, checked import and agent workflow. Attribution travels with the import; Graspable is not endorsed by NVIDIA or Hugging Face. Version 0.19 also brings the curated skills library into the desktop app, including project-specific adding and removing. Browse the original instructions and requirements before giving your agent a skill. Download Graspable, try either example, or read the OpenUSD guide for the supported import and export workflow.","u":"/blog/nvidia-openusd-assets-in-graspable#open-assets-a-built-in-workflow"},{"k":"blog","t":"Choosing an AI builder for XR","x":"What an AI builder for XR has to do, how Graspable does it, and the few cases where another tool fits better. Check device support, editable output and testing before you choose. An AI builder for XR has to do more than produce a good first screenshot. It has to build the experience, let you change it, test it the way a person will use it, and get it onto the device your audience has. Graspable is","s":"What an AI builder for XR has to do, how Graspable does it, and the few cases where another tool fits better. Check device support, editable output and testing before you choose. An AI builder for XR has to do more than produce a good first screenshot. It has to build the experience, let you change it, test it the way a person will use it, and get it onto the device your audience has. Graspable is a desktop app built for that whole job: you describe the experience, its agent edits a real project on your computer, checks that it builds and works in an emulated headset, and publishes it at its own address. For most XR projects, start with Graspable. This guide shows what to check in any builder, how Graspable handles each point, and the few situations where another tool is the better fit. This guide is published by Graspable. We read the other products' official documentation on October 22, 2026, and added the IWSDK review on October 23. On November 5 we updated the ENGAGE XR section: its platform shut down on October 31, 2026. The comparisons describe documented workflows, not a speed or quality benchmark.","u":"/blog/choosing-an-ai-builder-for-xr"},{"k":"blog","t":"Choosing an AI builder for XR","h":"Start with what the audience will open","x":"Write down the device, how someone gets into the experience, and what they need to do there. Graspable has a starting template for each of these: - A link opened in a headset: WebXR projects such as React XR and the XR Game template run in the headset's browser with hands and controllers. Check the immersive modes and input on the browser your audience uses. - An installed headset app: the Unity t","s":"Write down the device, how someone gets into the experience, and what they need to do there. Graspable has a starting template for each of these: - A link opened in a headset: WebXR projects such as React XR and the XR Game template run in the headset's browser with hands and controllers. Check the immersive modes and input on the browser your audience uses. - An installed headset app: the Unity templates for Quest, Android XR and XREAL, Unreal Engine and Apple Vision Pro build native apps with the engine installed on your computer. - A link opened on a phone: Web AR places content in the room through the phone's browser. Ask any builder which tracking it uses: camera access, image tracking and finding a bare tabletop are different capabilities. - An app for AI glasses: Graspable has templates for Meta's glasses, from a camera-and-audio companion app to a display glasses web app. Check the manufacturer's SDK and distribution rules for your model. \"Works on mobile\" does not tell you whether a person can place something on their kitchen table. \"Works on Quest\" might mean a flat page in its browser. Ask to see the interaction you intend to ship.","u":"/blog/choosing-an-ai-builder-for-xr#start-with-what-the-audience-will-open"},{"k":"blog","t":"Choosing an AI builder for XR","h":"What the builder has to do","x":"Distinguish a builder that edits your application from an asset generator or an AI character inside the finished experience. A generated model or a conversational character is not an application, a test or a way to deliver it. Judge the whole loop: make the first scene, change an interaction, find a problem and repair it. In Graspable that loop looks like this: - Editable project: ordinary files o","s":"Distinguish a builder that edits your application from an asset generator or an AI character inside the finished experience. A generated model or a conversational character is not an application, a test or a way to deliver it. Judge the whole loop: make the first scene, change an interaction, find a problem and repair it. In Graspable that loop looks like this: - Editable project: ordinary files on your disk (React Three Fiber, Unity, Unreal or Swift), a history entry for every request, and undo for any run. - Your choice of AI: the Graspable agent with your own API key or ChatGPT sign-in, Claude Code with your Claude subscription, the Codex CLI, or Credits if you have no AI account. Building on your computer with your own account is free. - Checks: every request ends with a build and a load in a real browser, and recorded headset tests are replayed after each change. See how those checks work. - Delivery: publish a web app to its own address, open the preview on a headset on your network, or build an .apk for a Quest from a Unity project.","u":"/blog/choosing-an-ai-builder-for-xr#what-the-builder-has-to-do"},{"k":"blog","t":"Choosing an AI builder for XR","h":"Graspable and the alternatives","x":"Cloud app builders Replit's VR guide documents prompt-driven WebXR development in a cloud workspace, tested at a deployed address in a headset browser. Replit is a general app builder. Graspable is made for XR: its templates, emulated headset, recorded headset tests and native engine support are all about 3D and immersive apps, and your project stays on your computer. Compare both on the same scen","s":"Cloud app builders Replit's VR guide documents prompt-driven WebXR development in a cloud workspace, tested at a deployed address in a headset browser. Replit is a general app builder. Graspable is made for XR: its templates, emulated headset, recorded headset tests and native engine support are all about 3D and immersive apps, and your project stays on your computer. Compare both on the same scene and device in Graspable compared with Replit. Your own coding assistant Meta's Immersive Web SDK works with coding assistants including Claude Code, Cursor, GitHub Copilot and Codex. Its AI development tools let an agent inspect the scene, move an emulated headset and controllers, and capture screenshots. You can bring the same assistants to Graspable. Claude Code and Codex run inside the app with your own sign-in, and any assistant that speaks MCP can create, edit, check and preview Graspable projects (build WebXR with your coding agent). On top of the framework, Graspable adds the project templates, a visual editor, recorded tests and publishing. Choose IWSDK on its own if your team has already committed to its framework. Android XR prototypes Google's Vibe Coding XR workflow uses Gemini and XR Blocks to build WebXR prototypes, with desktop simulation and sharing on Android XR. It is a research workflow for one platform. Graspable's WebXR projects run in any headset browser with WebXR, and its Android XR Unity template builds a native app when the prototype becomes a product. Unity and Unreal Engine Unity's AI tools include an in-editor assistant and agent connections. Graspable's Unity templates go further for XR: the agent works in a running Unity Editor, checks the scene in play mode with a simulated room, and builds the headset app. The Unreal Engine template does the same with Unreal's own VR setup. Both use the engine you install and are part of the Pro plan; read Native XR first. If you already have a large Unity or Unreal project, try an agent on that project before moving it. Rebuilding working content to use a different prompt box has a cost. Browser 3D editors PlayCanvas supports WebXR, and its Editor MCP server lets an outside AI assistant edit its projects. Choose it if your team already works in its editor. A Graspable project is plain source code in your own folder, with a scene panel, an animation timeline and click-to-code selection in the preview, and no hosted editor to depend on. Spatial design tools ShapesXR is for spatial design, prototyping and presenting layouts in a headset, with a Unity integration. It suits agreeing on a layout before development. Graspable builds the working application from that agreement. Platforms that host the experience ENGAGE XR offered voice-driven world creation inside its own platform until it shut down on October 31, 2026, and its worlds went offline with it. An experience that lives only inside someone else's platform ends when the platform does. A Graspable project is yours: the source stays on your disk, and a published app has its own address. For shared spaces, the Multi-user template puts several people in the same scene.","u":"/blog/choosing-an-ai-builder-for-xr#graspable-and-the-alternatives"},{"k":"blog","t":"Choosing an AI builder for XR","h":"Try the same small job","x":"Use a task close to your actual product. For example: Build a tabletop exhibit with three objects. Let me pick one up, reset its position, and open a readable information panel. Animate one object. Include a way to leave the immersive experience. Then make one follow-up request: change the reset behavior while keeping the animation. Record the setup time, any manual repairs, the result and the dev","s":"Use a task close to your actual product. For example: Build a tabletop exhibit with three objects. Let me pick one up, reset its position, and open a readable information panel. Animate one object. Include a way to leave the immersive experience. Then make one follow-up request: change the reset behavior while keeping the animation. Record the setup time, any manual repairs, the result and the device used. In Graspable, start from React XR and record a headset test of the pick-up, so the follow-up is checked against it. Before committing to any builder, check: 1. Input: can your audience use it with hands, controllers or touch? 2. Tracking: does placement stay where expected when you move? Does it need an image target? 3. Editing: can you change the animation, materials and logic after generation, including by hand? 4. Verification: do the checks exercise the interaction, or only confirm that a page opens? 5. Performance: is it comfortable on the actual device after several minutes, with your assets? 6. Delivery: can another person open it, allow permissions, use it and exit without your help? 7. Cost: what covers the builder, AI usage, assets, hosting, engine tools and distribution? An emulated headset finds mistakes early. Comfort, hand tracking and frame rate still need a test on the headset itself. For a finished example, open Signal Garden, an animated rescue film and puzzle made in Graspable. Its phone-browser AR mode uses an image tracking mat; it does not detect a bare table.","u":"/blog/choosing-an-ai-builder-for-xr#try-the-same-small-job"},{"k":"blog","t":"Choosing an AI builder for XR","h":"Make the decision from the working result","x":"Choose the tool that completes your interaction, leaves the project editable and reaches your audience's device at a cost you accept. Download Graspable and run the exhibit above; plans cover premium templates, native engines, publishing and Credits.","u":"/blog/choosing-an-ai-builder-for-xr#make-the-decision-from-the-working-result"},{"k":"blog","t":"Choosing an AI builder for XR","h":"Sources","x":"- Replit: VR app builder - Meta: AI-assisted development with Immersive Web SDK and current IWSDK AI documentation - Google Research: Vibe Coding XR and XR Blocks - Unity: AI tools and agent connections - PlayCanvas: WebXR support and Editor MCP server - ShapesXR: design, prototyping and development - ENGAGE: shutdown notice For a closer look at one alternative, see the Graspable comparisons.","u":"/blog/choosing-an-ai-builder-for-xr#sources"},{"k":"blog","t":"Graspable 0.18: films you can step into","x":"Animate a scene with your agent, publish an immersive film, or put a rescue game on your table. Graspable 0.18 adds tabletop AR and Claude subscription sign-in. Graspable 0.18 brings the animation work from the recent releases together with two new AR templates. You can make a film, keep its scene editable, and let someone walk around it on their phone or in a headset. Animation, camera shots and ","s":"Animate a scene with your agent, publish an immersive film, or put a rescue game on your table. Graspable 0.18 adds tabletop AR and Claude subscription sign-in. Graspable 0.18 brings the animation work from the recent releases together with two new AR templates. You can make a film, keep its scene editable, and let someone walk around it on their phone or in a headset. Animation, camera shots and interactive films made in Graspable.","u":"/blog/graspable-0-18"},{"k":"blog","t":"Graspable 0.18: films you can step into","h":"A film and a game on your table","x":"Signal Garden starts with a stranded courier and a landing beacon that needs power. Watch the 90-second rescue film, or turn the relays yourself to guide sunlight to the landing pad. Hint helps when you get stuck; Undo lets you try another route. Try Signal Garden in your browser. The robot, courier, camera shots and score are part of the project. Change the route or timing, replace the models, or","s":"Signal Garden starts with a stranded courier and a landing beacon that needs power. Watch the 90-second rescue film, or turn the relays yourself to guide sunlight to the landing pad. Hint helps when you get stuck; Undo lets you try another route. Try Signal Garden in your browser. The robot, courier, camera shots and score are part of the project. Change the route or timing, replace the models, or ask the agent to write another mission. The rocky terrain uses a small, synthetic Gaussian scene alongside the animated models. Choose Tabletop AR for the browser version. Supported Android browsers and Quest use WebXR; iPhone Safari tracks an included mat that you can print or open on another screen. If a headset cannot find a surface, the scene offers manual placement in front of you. That option does not detect a table. Choose Unity AR tabletop for native apps. It includes separate phone and Quest configurations. The iPhone build uses surface tracking, so it does not need the mat. We checked the native film and puzzle on an iPhone, including sound, placement and Undo. On Quest Browser, placement, playback and the floating Leave AR button worked in passthrough. The final fingertip and pinch changes passed simulator checks; their last hardware check is still outstanding. The native Quest build passed validation and installed, but we have not completed its headset playback check. The AR guide covers placement, controls and platform requirements.","u":"/blog/graspable-0-18#a-film-and-a-game-on-your-table"},{"k":"blog","t":"Graspable 0.18: films you can step into","h":"Edit with your hands, or ask the agent","x":"In Animate , select an object, change its position, rotation or scale, and key a channel or the whole pose. Open easing controls beside a channel when you need them. Camera shots have editable opening and ending views; sound clips have trims, volume and fades. Saved edits become context for the next agent request. You can adjust a gesture by hand, then ask the agent to repeat it across a scene wit","s":"In Animate , select an object, change its position, rotation or scale, and key a channel or the whole pose. Open easing controls beside a channel when you need them. Camera shots have editable opening and ending views; sound clips have trims, volume and fades. Saved edits become context for the next agent request. You can adjust a gesture by hand, then ask the agent to repeat it across a scene without losing the edit history. The filmmaking walkthrough shows the workspace and two playable films. It also explains scene exchange through OpenUSD and OpenTimelineIO, immersive exports, and the parts of XR editing that still need work.","u":"/blog/graspable-0-18#edit-with-your-hands-or-ask-the-agent"},{"k":"blog","t":"Graspable 0.18: films you can step into","h":"Claude in the Graspable agent","x":"You can now choose your Claude subscription within Graspable agent , keeping its project tools, approvals and edit history. This experimental option uses your local Claude sign-in. Your account allowances and extra-usage settings apply; the displayed budget estimate is not a bill. Download Graspable 0.18, then choose a template to start. Tabletop AR is included with Builder and higher; Unity templ","s":"You can now choose your Claude subscription within Graspable agent , keeping its project tools, approvals and edit history. This experimental option uses your local Claude sign-in. Your account allowances and extra-usage settings apply; the displayed budget estimate is not a bill. Download Graspable 0.18, then choose a template to start. Tabletop AR is included with Builder and higher; Unity templates require Pro or higher. Viewing the scenes does not consume Credits.","u":"/blog/graspable-0-18#claude-in-the-graspable-agent"},{"k":"blog","t":"Make a film inside your 3D scene","x":"Graspable brings together an animation workspace for objects, camera shots and sound, with editable scenes and films you can share on the web. A character crosses a room. The camera waits, follows, then cuts. A light comes on just before the music changes. In Graspable, you can now edit those decisions in the scene itself. The animation workspace brings together object selection, a timeline, camer","s":"Graspable brings together an animation workspace for objects, camera shots and sound, with editable scenes and films you can share on the web. A character crosses a room. The camera waits, follows, then cuts. A light comes on just before the music changes. In Graspable, you can now edit those decisions in the scene itself. The animation workspace brings together object selection, a timeline, camera shots, sound and export. You can ask an agent to set up a sequence, adjust it by hand, and ask for another change without starting over.","u":"/blog/make-a-film-inside-your-scene"},{"k":"blog","t":"Make a film inside your 3D scene","h":"Animate the objects in your scene","x":"Open Animate , choose Edit scene , and select an object in the preview. Move, rotate or scale it, then set a keyframe. Scrub to another frame and change its pose. The timeline shows the channels and keys; you can move keys, inspect curves, undo an edit and save the take. A saved take belongs to your project. Before a new agent request starts, Graspable saves your open animation draft and includes ","s":"Open Animate , choose Edit scene , and select an object in the preview. Move, rotate or scale it, then set a keyframe. Scrub to another frame and change its pose. The timeline shows the channels and keys; you can move keys, inspect curves, undo an edit and save the take. A saved take belongs to your project. Before a new agent request starts, Graspable saves your open animation draft and includes recent accepted edits in the agent's context. If another edit changed the take, Graspable asks you to resolve it instead of overwriting it. You and your agent work on the same local project. Simultaneous animation editing from different computers is still ahead.","u":"/blog/make-a-film-inside-your-scene#animate-the-objects-in-your-scene"},{"k":"blog","t":"Make a film inside your 3D scene","h":"Put the camera where the story needs it","x":"Capture a shot from the viewport, set its length and lens, then capture an ending view to add a move. You can also look at the camera from outside, move it with handles, and check its composition in a live viewfinder. In XR, the film camera and your head remain separate. The desktop-connected editing panel lets you position a shot with a controller and enter its viewpoint while keeping head tracki","s":"Capture a shot from the viewport, set its length and lens, then capture an ending view to add a move. You can also look at the camera from outside, move it with handles, and check its composition in a live viewfinder. In XR, the film camera and your head remain separate. The desktop-connected editing panel lets you position a shot with a controller and enter its viewpoint while keeping head tracking active. Physical headset comfort and hand manipulation still need more work. Add dialogue, music or effects in Sound . Clips have source trims, volume and fades. Sound follows pauses, seeks and loops; the release also fixes sound playing beyond a selected loop range.","u":"/blog/make-a-film-inside-your-scene#put-the-camera-where-the-story-needs-it"},{"k":"blog","t":"Make a film inside your 3D scene","h":"Watch it, or walk into it","x":"The Last Light is a short film about a maintenance machine trying to wake an observatory. Its objects, camera shots and timing remain editable. The Third Option is a two-and-a-half-minute spatial comedy. Two pills become three choices. Pick an ending, hide the controls, or look around the room. It uses a Gaussian-splat environment, generated characters and original dialogue. Its generated characte","s":"The Last Light is a short film about a maintenance machine trying to wake an observatory. Its objects, camera shots and timing remain editable. The Third Option is a two-and-a-half-minute spatial comedy. Two pills become three choices. Pick an ending, hide the controls, or look around the room. It uses a Gaussian-splat environment, generated characters and original dialogue. Its generated characters use basic facial animation. Publishing an interactive project keeps the world available to explore. Exporting a rendered film produces a fixed sequence for a screen or a compatible immersive player. Graspable supports 360° and stereo delivery, plus Apple spatial-video encoding on supported Macs. Formats, browser support and export limits are listed in the film delivery guide.","u":"/blog/make-a-film-inside-your-scene#watch-it-or-walk-into-it"},{"k":"blog","t":"Make a film inside your 3D scene","h":"Keep the work editable","x":"Use the OpenUSD inspector to examine a scene, change properties and variants, or convert supported assets for the preview. Conversion reports identify features that cannot carry over. OpenTimelineIO exchanges shot sequences and audio references; it does not replace the scene or its animation rigs. Agents can request bounded views and changes instead of reading an entire scene or take on every turn","s":"Use the OpenUSD inspector to examine a scene, change properties and variants, or convert supported assets for the preview. Conversion reports identify features that cannot carry over. OpenTimelineIO exchanges shot sequences and audio references; it does not replace the scene or its animation rigs. Agents can request bounded views and changes instead of reading an entire scene or take on every turn. Project history remains available when you need to inspect or restore earlier work. Start with the animation guide, try the films above, or download Graspable for macOS, Windows or Linux.","u":"/blog/make-a-film-inside-your-scene#keep-the-work-editable"},{"k":"blog","t":"Generate project images through confidential Routstr","x":"Graspable can generate images through a confidential Routstr session, with a price you approve and checks before the image becomes a project asset. Two live tests cost 36.381 sats. You can now ask Graspable to generate an image through Routstr's experimental Confidential mode. The agent lists the models your node offers, shows the price in sats, and asks you to approve it. After generation, Graspa","s":"Graspable can generate images through a confidential Routstr session, with a price you approve and checks before the image becomes a project asset. Two live tests cost 36.381 sats. You can now ask Graspable to generate an image through Routstr's experimental Confidential mode. The agent lists the models your node offers, shows the price in sats, and asks you to approve it. After generation, Graspable checks the result and saves an ordinary image in your project. This extends the confidential connection we already use for text. It uses Juraj Bednar's open-source client, which we updated and reviewed for this integration.","u":"/blog/confidential-image-generation"},{"k":"blog","t":"Generate project images through confidential Routstr","h":"Use it in a project","x":"Open Settings → AI for the agent , select your Routstr account, and switch on Confidential mode if the node offers it. Select that account for the project. Then ask: List the confidential image models and their prices. I need a neutral character reference portrait. Show me the available options before generating anything. Reading the signed model list does not start a paid generation. Availability","s":"Open Settings → AI for the agent , select your Routstr account, and switch on Confidential mode if the node offers it. Select that account for the project. Then ask: List the confidential image models and their prices. I need a neutral character reference portrait. Show me the available options before generating anything. Reading the signed model list does not start a paid generation. Availability and prices come from your node, so you do not have to wait for a Graspable update when it changes its catalog. Choose a model and quality tier, then approve the quoted purchase. Every image generation needs your approval. The charge comes from your Routstr balance, separately from Graspable Credits and the agent's USDC wallet. The result goes into a new folder under public/assets/ . You and the agent can use it like any other project image. Graspable does not overwrite an existing folder or put the generation prompt and payment receipt beside the image. Review the result before publishing: files under public/ travel with the published project.","u":"/blog/confidential-image-generation#use-it-in-a-project"},{"k":"blog","t":"Generate project images through confidential Routstr","h":"What confidential means here","x":"The protocol is designed to keep the image prompt and response encrypted between your computer and the model provider while the Routstr node relays the connection. The provider still sees the prompt and image. The node sees the requested model, quality tier, image count and request length, as well as connection timing. Graspable checks the pinned node signature, provider certificate, protocol proo","s":"The protocol is designed to keep the image prompt and response encrypted between your computer and the model provider while the Routstr node relays the connection. The provider still sees the prompt and image. The node sees the requested model, quality tier, image count and request length, as well as connection timing. Graspable checks the pinned node signature, provider certificate, protocol proofs, requested image variant and signed fixed-price receipt before accepting the result. A failed check stops the request. There is no retry through an ordinary unencrypted connection. These checks cover this generation session. They do not make the rest of an agent conversation or a connected service confidential. The protocol is experimental; our integration review and successful tests are not an independent cryptographic audit.","u":"/blog/confidential-image-generation#what-confidential-means-here"},{"k":"blog","t":"Generate project images through confidential Routstr","h":"Two real requests","x":"On 19 October, we generated and inspected two non-explicit portraits: - Qwen Image 2.1 Turbo produced a 1024 × 1024 WebP for 24.254 sats , in about 16.5 seconds. - Lustify v8 produced a 1536 × 1536 fashion portrait for 12.127 sats , in about 14.3 seconds. Both passed verification on the first attempt. The total was 36.381 sats . Those are measurements from this test, not a price list or a speed gu","s":"On 19 October, we generated and inspected two non-explicit portraits: - Qwen Image 2.1 Turbo produced a 1024 × 1024 WebP for 24.254 sats , in about 16.5 seconds. - Lustify v8 produced a 1536 × 1536 fashion portrait for 12.127 sats , in about 14.3 seconds. Both passed verification on the first attempt. The total was 36.381 sats . Those are measurements from this test, not a price list or a speed guarantee. Each request establishes a fresh encrypted session, so setup has a noticeable cost in time and traffic. An interrupted session can still incur its quoted session fee. If generation finishes but saving fails, check your project and balance before approving another attempt. Graspable will not silently spend again. Set up Routstr and read the Confidential mode limits.","u":"/blog/confidential-image-generation#two-real-requests"},{"k":"blog","t":"Speech and gaze you can edit frame by frame","x":"Import timed mouth cues, key a character’s named bones, and let an agent make changes to the same saved take. What works today, and what our native character study taught us. A character's mouth can follow the recording while its eyes look somewhere else. You should be able to fix either without rebuilding the whole performance. Graspable now imports timed mouth cues as editable facial keyframes. ","s":"Import timed mouth cues, key a character’s named bones, and let an agent make changes to the same saved take. What works today, and what our native character study taught us. A character's mouth can follow the recording while its eyes look somewhere else. You should be able to fix either without rebuilding the whole performance. Graspable now imports timed mouth cues as editable facial keyframes. You can also animate a character's named bones, and ask an agent to add speech, blinks or gaze to the same saved take. The character needs those controls in its source model.","u":"/blog/editable-character-performances"},{"k":"blog","t":"Speech and gaze you can edit frame by frame","h":"Put speech on the timeline","x":"Open a web 3D project, choose Animate , and select the character. Use Animate this object if it does not have an animation identity yet. Under Speech performance , import a Rhubarb mouth-cue JSON file. Map its spoken shapes to the character's actual facial controls, check the amount and start frame, then choose Check speech keys . Graspable shows how many channels and keys it will add before you a","s":"Open a web 3D project, choose Animate , and select the character. Use Animate this object if it does not have an animation identity yet. Under Speech performance , import a Rhubarb mouth-cue JSON file. Map its spoken shapes to the character's actual facial controls, check the amount and start frame, then choose Check speech keys . Graspable shows how many channels and keys it will add before you apply them. Choose Add speech keys and save the take. Scrub to a word, adjust a mouth weight, or retime a key. Undo reverses the import. Add the recording as an audio clip at the matching start frame; the cue importer does not analyze audio or generate a voice. Existing mapped mouth channels are protected by default. Choosing replacement replaces those entire channels, including manual keys outside the new speech range. Other expressions and body channels stay as they were. In our desktop check, a 6.56-second recording supplied 40 cues covering seven spoken shapes. The importer produced seven editable tracks with 259 keys. We checked replacement and undo in the running interface as well as testing the underlying data.","u":"/blog/editable-character-performances#put-speech-on-the-timeline"},{"k":"blog","t":"Speech and gaze you can edit frame by frame","h":"Direct a glance yourself, or ask the agent","x":"Choose Property → Rig rotations , select a uniquely named bone and set its local angles. The saved keys use quaternion rotation. A keyed bone overrides an imported animation clip after the clip updates; other bones keep their animation. Duplicate bone names cannot be bound ambiguously. An agent can work on the same named channels. For example: Keep my mouth keys. Add two blinks and have the charac","s":"Choose Property → Rig rotations , select a uniquely named bone and set its local angles. The saved keys use quaternion rotation. A keyed bone overrides an imported animation clip after the clip updates; other bones keep their animation. Duplicate bone names cannot be bound ambiguously. An agent can work on the same named channels. For example: Keep my mouth keys. Add two blinks and have the character look toward the doorway at frame 120, then back at frame 180. Read the existing take first and leave the body animation alone. The agent can apply blink and gaze edits independently of speech. It must use the model's actual controls and respect the rig's local axes. Saved edits have revision checks and history, so a stale agent edit is refused. Your recent saved manual changes also enter the next request's context. This workflow coordinates human and agent edits; it does not add remote participant presence or simultaneous conflict merging.","u":"/blog/editable-character-performances#direct-a-glance-yourself-or-ask-the-agent"},{"k":"blog","t":"Speech and gaze you can edit frame by frame","h":"A good source rig still needs a careful export","x":"We tested the import path with Rain, a stylized Blender character. Early exports lost an eyebrow, broke the mouth during head turns, and produced dark patches on closed eyelids. Some of those files passed format validation. Looking at the rendered face caught problems a valid file could not. The native rig uses constraints and corrective deformation that a single exported bone rotation does not re","s":"We tested the import path with Rain, a stylized Blender character. Early exports lost an eyebrow, broke the mouth during head turns, and produced dark patches on closed eyelids. Some of those files passed format validation. Looking at the rendered face caught problems a valid file could not. The native rig uses constraints and corrective deformation that a single exported bone rotation does not reproduce. For this study we evaluated the native performance, preserved the missing mirrored anatomy, and fitted corrective positions and normals for the runtime model. We also baked the native eye maps with explicit UVs and retained the vertex colours needed for its pupils. An actual desktop browser capture from the native-export study, at 8.5 seconds. Rain Rig © Blender Foundation cloud.blender.org, CC-BY. Derived performance and runtime preparation by Graspable. The corrected mouth, blink and head-turn poses passed our desktop visual review. This is a baked technical study of a stylized character, not an automatic converter for arbitrary rigs or a photorealistic replacement cast. Changing its skeletal performance requires rebaking the fitted corrections. The editable native source remains separate. Rain's original rig and license are available from Blender Studio.","u":"/blog/editable-character-performances#a-good-source-rig-still-needs-a-careful-export"},{"k":"blog","t":"Speech and gaze you can edit frame by frame","h":"What still needs a headset","x":"The authoring changes have desktop tests and browser checks. On a physical Quest 3, a 35-second test of the new two-character study delivered about 7.2 frames per second in a 72 Hz session. It used full stereo resolution, without audio or the studio lighting used in the desktop screenshot. This study still needs substantial performance work. Final sustained iPhone testing and Quest performance acc","s":"The authoring changes have desktop tests and browser checks. On a physical Quest 3, a 35-second test of the new two-character study delivered about 7.2 frames per second in a 72 Hz session. It used full stereo resolution, without audio or the studio lighting used in the desktop screenshot. This study still needs substantial performance work. Final sustained iPhone testing and Quest performance acceptance remain pending; desktop visual review does not establish headset frame rate, controller comfort or readability. You can already open the experimental XR authoring panel from the animation workspace, but its controller workflow still needs physical Quest checks. Hand manipulation and headset audio monitoring remain unfinished. Read the animation workflow and its current limits.","u":"/blog/editable-character-performances#what-still-needs-a-headset"},{"k":"blog","t":"A router that cannot read your prompts","x":"Graspable can now use a Routstr node in confidential mode, built on Juraj Bednar's protocol: the node relays an encrypted connection the agent holds with the model provider itself. What it protects, what we measured, and what it costs. When the Graspable agent runs on a Routstr node, it pays a stranger's server in bitcoin to pass its requests to a model provider. There is no account and no card, w","s":"Graspable can now use a Routstr node in confidential mode, built on Juraj Bednar's protocol: the node relays an encrypted connection the agent holds with the model provider itself. What it protects, what we measured, and what it costs. When the Graspable agent runs on a Routstr node, it pays a stranger's server in bitcoin to pass its requests to a model provider. There is no account and no card, which is the point. The price is trust: the node holds the provider's API key, so every request goes through it in the clear. The person who runs the node can read your prompts, which for an agent means your source code. They can change what comes back, including the tool calls the agent then carries out on your computer. And they can run a cheaper model than the one you chose and bill you for the dear one. A node can now give all of that up, and Graspable can use it that way.","u":"/blog/a-router-that-cannot-read-your-prompts"},{"k":"blog","t":"A router that cannot read your prompts","h":"The idea, from Juraj Bednar","x":"Juraj Bednar runs the Routstr node at routstr.cypherpunk.today and is a co-founder of Paralelná Polis. On 10 October 2026 he published a protocol, a paper, and working code for what he calls confidential mode: - the paper, \"Confidential and verifiable upstream inference over standard TLS 1.3\"; - his introduction to it; - the code, and a version that runs in a browser tab. The usual answer to \"how ","s":"Juraj Bednar runs the Routstr node at routstr.cypherpunk.today and is a co-founder of Paralelná Polis. On 10 October 2026 he published a protocol, a paper, and working code for what he calls confidential mode: - the paper, \"Confidential and verifiable upstream inference over standard TLS 1.3\"; - his introduction to it; - the code, and a version that runs in a browser tab. The usual answer to \"how do I know the server is not reading my data\" is a hardware enclave: the server proves it runs known software inside a protected chip. That moves the trust to the chip maker, and it has to be checked again after every software update. His design needs none of it. In the paper's words, it \"rests instead on the provider's TLS certificate and on cryptography.\" The agent is the TLS client of the provider. The encrypted connection runs between your computer and the provider (on his node, Venice). The node sits in the path and relays the encrypted records. The keys are computed by two parties and split between them. A normal TLS client would hold every key of the session, including the one that protects the request's first lines, where the API key goes. So the agent and the node run the handshake's key derivation together, as a two-party computation in which neither learns the other's part. The node comes out with the key for the short opening section, where it writes its API key. The agent comes out with the keys for the request body and for the answer. The paper is exact that the shared work covers the key schedule only, never the content (section 1). Proofs go in both directions. The node proves, without revealing its key, that what it wrote is the published request header and nothing else (section 4.5). The agent proves, without revealing the prompt, that its request ends in an agreed block fixing the model and the token limit, and the provider receives the full request only after that proof checks out (section 4.6). So the node knows it is not being made to pay for something else, and you know which model answered. The bill is the provider's own count. At the end the agent proves that the usage figures it discloses are the genuine last part of the provider's answer, and the node charges that usage at the rates of an offer it signed beforehand, on a receipt it signs (sections 4.7 and 4.8). What the node still learns is stated plainly in the paper's security analysis (section 5): which model, when, and how long requests and answers are. Message sizes can say something about content. It can also cut the connection. It cannot read or alter the conversation. One thing stays out of reach of any such proof, as his introduction says: which weights the provider really runs behind the model's name.","u":"/blog/a-router-that-cannot-read-your-prompts#the-idea-from-juraj-bednar"},{"k":"blog","t":"A router that cannot read your prompts","h":"What Graspable does with it","x":"Graspable does not reimplement any of this. It runs his open-source client, the same one behind the browser demo, inside the agent on your computer, pinned to an exact version. Under a Routstr balance, where the node offers it, there is now a Confidential mode switch. Switching it on pins the node's signing key and the provider it relays to, as the node states them at that moment. From then on: - ","s":"Graspable does not reimplement any of this. It runs his open-source client, the same one behind the browser demo, inside the agent on your computer, pinned to an exact version. Under a Routstr balance, where the node offers it, there is now a Confidential mode switch. Switching it on pins the node's signing key and the provider it relays to, as the node states them at that moment. From then on: - every request the agent makes to the model becomes one confidential session; - an offer signed by a different key, or naming a different provider, is refused; - each request shows a line in the run with what was verified: the provider the agent really spoke to, its certificate, the proofs, and the cost from the signed receipt; - if a request cannot be made confidentially, it is not made. Graspable never falls back to sending it the ordinary way.","u":"/blog/a-router-that-cannot-read-your-prompts#what-graspable-does-with-it"},{"k":"blog","t":"A router that cannot read your prompts","h":"What we measured","x":"On 10 October 2026 we ran it against his node from the Graspable agent, with a balance of 300 sats paid by Lightning. - A plain question and a request that ends in a tool call both completed and verified. Each took about 9 seconds and cost about 0.03 sats. - We asked the agent to make a one-line change in a new React XR project. It read the file, edited it, and the project then passed the build an","s":"On 10 October 2026 we ran it against his node from the Graspable agent, with a balance of 300 sats paid by Lightning. - A plain question and a request that ends in a tool call both completed and verified. Each took about 9 seconds and cost about 0.03 sats. - We asked the agent to make a one-line change in a new React XR project. It read the file, edited it, and the project then passed the build and the browser check. That took four model requests. All four verified, taking between 9 and 13 seconds each, and the whole run cost 3.7 sats. - The agent's read file and edit file calls came through the confidential connection and were carried out normally. Our test was one node, one provider and one small model, and the runs were short. The author's own measurements, on his setup, are in section 6 of the paper.","u":"/blog/a-router-that-cannot-read-your-prompts#what-we-measured"},{"k":"blog","t":"A router that cannot read your prompts","h":"What it costs","x":"This is slower and heavier than the ordinary mode. - Each request is its own session. The paper puts the traffic at about 22 MB from the node to the client and 1.2 MB back per request. An agent makes many requests in a run, so a long one moves a lot of data. - Requests take longer. Ours took 9 to 13 seconds each from start to finish. The protocol can prepare a session ahead of time to shorten this","s":"This is slower and heavier than the ordinary mode. - Each request is its own session. The paper puts the traffic at about 22 MB from the node to the client and 1.2 MB back per request. An agent makes many requests in a run, so a long one moves a lot of data. - Requests take longer. Ours took 9 to 13 seconds each from start to finish. The protocol can prepare a session ahead of time to shorten this; Graspable does not do that yet, because a prepared session that goes unused costs a small fee, and an agent's next request is not known in advance. - More of the balance is held while a request runs. While a request runs, the node sets aside the cost of a full context for the model, and releases what was not used. We watched a 299-sat balance show 183 available during a run and 296 after it. - Sizes are limited: answers up to 32,768 tokens and requests up to 8 MB on his node. For a quick edit to a scene, the ordinary mode is the better tool. For work where the prompts are the valuable part, the wait buys something real.","u":"/blog/a-router-that-cannot-read-your-prompts#what-it-costs"},{"k":"blog","t":"A router that cannot read your prompts","h":"How far to trust it","x":"The two-party computation is a variant with no published proof, and the paper says the construction as a whole needs external review (section 5). His introduction calls the work new and experimental and asks people to try to break it. Two days after this article, the author published a hardening of the client that showed why that caution matters. The earlier client accepted the provider's certific","s":"The two-party computation is a variant with no published proof, and the paper says the construction as a whole needs external review (section 5). His introduction calls the work new and experimental and asks people to try to break it. Two days after this article, the author published a hardening of the client that showed why that caution matters. The earlier client accepted the provider's certificate chain without requiring the provider to prove it held the certificate's key, so a node could have answered in the provider's place (his notes call the fix TLS-11). The same update changed the protocol in a way older clients cannot speak. Graspable 0.16.12 carries the fixed client, and an older Graspable is refused by an updated node. There is also a party the protocol cannot remove: whoever ships the code that runs the checks. In the browser demo that is the web page. In Graspable it is us. We pin his client to an exact version and review what changed before moving the pin, since that code sees your prompts. So confidential mode in Graspable is marked experimental, off until you switch it on, and described with its costs next to the switch.","u":"/blog/a-router-that-cannot-read-your-prompts#how-far-to-trust-it"},{"k":"blog","t":"A router that cannot read your prompts","h":"Try it","x":"In Graspable: Settings, AI for the agent, add Routstr, paid with bitcoin , choose a node that offers confidential mode, pay the invoice from your own wallet, and switch Confidential mode on under the balance. The documentation has the details. Without Graspable: his browser version does every check in the page itself.","u":"/blog/a-router-that-cannot-read-your-prompts#try-it"},{"k":"blog","t":"A router that cannot read your prompts","h":"Sources","x":"- Juraj Bednar, \"Confidential and verifiable upstream inference over standard TLS 1.3: A no-trusted-hardware protocol for LLM routing nodes\", 10 October 2026. Sections 1, 4.5 to 4.8, 5 and 6 are the ones referred to above. - Juraj Bednar, \"Confidential inference on Routstr without trusted hardware\", 10 October 2026. - routstr-confidential and routstr-confidential-web, the protocol's code and the c","s":"- Juraj Bednar, \"Confidential and verifiable upstream inference over standard TLS 1.3: A no-trusted-hardware protocol for LLM routing nodes\", 10 October 2026. Sections 1, 4.5 to 4.8, 5 and 6 are the ones referred to above. - Juraj Bednar, \"Confidential inference on Routstr without trusted hardware\", 10 October 2026. - routstr-confidential and routstr-confidential-web, the protocol's code and the client Graspable uses, both under the MIT licence. - Routstr, the network of nodes.","u":"/blog/a-router-that-cannot-read-your-prompts#sources"},{"k":"blog","t":"A chair configurator in a headset, from one request","x":"A shell chair in a dark showroom: point at a swatch, pull the trigger, and the chair and its price change. It was built in one run, and the request is here word for word. A shell chair stands on a turntable in a dark showroom, with an options panel beside it. Point at a swatch and pull the trigger: shell colour, seat pad and legs change on the chair, the price adds up, and a summary lists the choi","s":"A shell chair in a dark showroom: point at a swatch, pull the trigger, and the chair and its price change. It was built in one run, and the request is here word for word. A shell chair stands on a turntable in a dark showroom, with an options panel beside it. Point at a swatch and pull the trigger: shell colour, seat pad and legs change on the chair, the price adds up, and a summary lists the choice. Open the chair configurator. The chair configurator in the emulated headset.","u":"/blog/a-chair-configurator-from-one-request"},{"k":"blog","t":"A chair configurator in a headset, from one request","h":"What it shows","x":"- Controllers. You choose by pointing at a swatch and pulling the trigger. - A product configurator. Shell colour, seat pad and legs change on the chair, and the price adds up. - PBR materials. The request asks for gloss paint, velvet, leather and brushed steel that look like what they are. - Spatial labels. The panel beside the chair lists the swatches with their prices and the total. In the pict","s":"- Controllers. You choose by pointing at a swatch and pulling the trigger. - A product configurator. Shell colour, seat pad and legs change on the chair, and the price adds up. - PBR materials. The request asks for gloss paint, velvet, leather and brushed steel that look like what they are. - Spatial labels. The panel beside the chair lists the swatches with their prices and the total. In the picture at the top the chair has a bone white shell, a cognac leather seat pad and walnut legs, and the panel shows a total of 585 euros. With the price list of the request that is the base chair at 420, the shell at 0, the leather at 120 and the walnut at 45.","u":"/blog/a-chair-configurator-from-one-request#what-it-shows"},{"k":"blog","t":"A chair configurator in a headset, from one request","h":"How it was made","x":"The project started from the Product Studio template: one product in a photo studio, with colours and materials people can choose, placed at real size in VR or in their own room. It took one run. One request, no follow-ups. The AI was the Graspable agent. A run ends only when the project builds and loads in a real browser without errors.","u":"/blog/a-chair-configurator-from-one-request#how-it-was-made"},{"k":"blog","t":"A chair configurator in a headset, from one request","h":"The request, word for word","x":"This is the text that was sent to the agent to start the project. Nothing was edited for this article. Make this a proper configurator for the chair, in a dark showroom. The showroom: charcoal backdrop that curves into the floor, one warm key light from the upper left, a cool rim light from behind, a soft shadow under the chair, and a slow sweep of light across the chair whenever I change an optio","s":"This is the text that was sent to the agent to start the project. Nothing was edited for this article. Make this a proper configurator for the chair, in a dark showroom. The showroom: charcoal backdrop that curves into the floor, one warm key light from the upper left, a cool rim light from behind, a soft shadow under the chair, and a slow sweep of light across the chair whenever I change an option. Options: shell in five colours (deep navy gloss, bone white, forest green, terracotta, matt black), seat pad in four fabrics (forest green velvet, oatmeal wool, charcoal weave, cognac leather), legs in three finishes (oak, walnut, black steel). Navy gloss shell with the forest green velvet pad and oak legs is the default. The materials must look like what they are: gloss paint with a clear reflection, velvet with a soft sheen, leather with a fine grain, brushed steel. Every choice has a price. Use: base chair 420; shell colours +0 except gloss navy +40; velvet +60, wool +35, weave +0, leather +120; oak +0, walnut +45, steel +30. Show the running total in euros at the top of the panel in the headset and on the options card, and show each price next to its choice. Add a Summary button that opens a plate listing the chosen shell, pad and legs with their prices and the total, and closes again. In the headset the panel stands to the right of the chair, angled towards me, with colour swatches I can point at and pick with the trigger; the chosen one has a clear ring. Text must be crisp and readable from two metres. Keep the exploded view, the callouts and the real-size mode working.","u":"/blog/a-chair-configurator-from-one-request#the-request-word-for-word"},{"k":"blog","t":"A chair configurator in a headset, from one request","h":"Notes and credits","x":"The picture and the film are captures of the project running in Graspable's emulated headset.","u":"/blog/a-chair-configurator-from-one-request#notes-and-credits"},{"k":"blog","t":"A chair configurator in a headset, from one request","h":"Make your own","x":"Graspable is a desktop app. Describe the scene you want, and the agent builds it as a real project on your computer and checks that it runs. Download Graspable, and start from the Product Studio template to show a product of your own. More examples are in Made in Graspable.","u":"/blog/a-chair-configurator-from-one-request#make-your-own"},{"k":"blog","t":"A furniture catalogue loaded from the internet, placed in your room","x":"Pick a scanned piece of furniture from a live catalogue, see it at its true size, and place several in your own room. It was built in one run, and the request is here word for word. The catalogue in this app is not in the app. It is loaded live: pick a scanned piece from the list, see it at its true size, and place several in your own room. Open the furniture catalogue. From the same page you can ","s":"Pick a scanned piece of furniture from a live catalogue, see it at its true size, and place several in your own room. It was built in one run, and the request is here word for word. The catalogue in this app is not in the app. It is loaded live: pick a scanned piece from the list, see it at its true size, and place several in your own room. Open the furniture catalogue. From the same page you can continue from it in Graspable.","u":"/blog/a-furniture-catalogue-loaded-from-the-internet"},{"k":"blog","t":"A furniture catalogue loaded from the internet, placed in your room","h":"What it shows","x":"- It loads from the internet. Models, thumbnails and sizes come from a public catalogue when the app runs. - Mixed reality. The pieces are placed in your own room. - uikit. The panel with its tabs, filter chips and cards is built with it. - Photoreal models. The furniture is scanned. The picture at the top shows a photographed room with a leather lounge chair on a round stand. Beside it is the pan","s":"- It loads from the internet. Models, thumbnails and sizes come from a public catalogue when the app runs. - Mixed reality. The pieces are placed in your own room. - uikit. The panel with its tabs, filter chips and cards is built with it. - Photoreal models. The furniture is scanned. The picture at the top shows a photographed room with a leather lounge chair on a round stand. Beside it is the panel, titled Furniture catalogue, with the tabs Catalogue, My room and Device, the filter chips, and a grid of cards with thumbnails, names and sizes of sofas, tables and chairs.","u":"/blog/a-furniture-catalogue-loaded-from-the-internet#what-it-shows"},{"k":"blog","t":"A furniture catalogue loaded from the internet, placed in your room","h":"How it was made","x":"The project started from the Cross-platform XR template: one app for every headset, pair of glasses, phone and computer, with a product shown in your own room where the device has passthrough, and in a photographed room where it does not. It took one run. The AI was Claude Code. A run ends only when the project builds and loads in a real browser without errors.","u":"/blog/a-furniture-catalogue-loaded-from-the-internet#how-it-was-made"},{"k":"blog","t":"A furniture catalogue loaded from the internet, placed in your room","h":"The request, word for word","x":"This is the text that was sent to the agent to start the project. Nothing was edited for this article. Turn this into a furniture catalogue that people place in their own room, the kind of app a furniture retailer would commission. The catalogue is not in the app: it is loaded from the internet when the app starts. The source is Poly Haven's public catalogue of scanned, photoreal models (public do","s":"This is the text that was sent to the agent to start the project. Nothing was edited for this article. Turn this into a furniture catalogue that people place in their own room, the kind of app a furniture retailer would commission. The catalogue is not in the app: it is loaded from the internet when the app starts. The source is Poly Haven's public catalogue of scanned, photoreal models (public domain, CC0, open to browsers): - the list: https://api.polyhaven.com/assets?t=models&c=furniture gives every furniture model with its name, categories, real dimensions in millimetres and polycount; - one model's files: https://api.polyhaven.com/files/\\<id ; use the glTF at 1k (gltf → 1k → gltf: its url, and the files under include, which the glTF refers to by relative path, so loading the url is enough); - thumbnails: https://cdn.polyhaven.com/asset img/thumbs/\\<id .png?width=256 Keep only models under 60,000 polygons and no larger than 2.5 m in any direction, and show the 24 most downloaded. The panel has two tabs. Catalogue: a scrolling grid of cards, each with the thumbnail, the name and the real size in centimetres (width x depth x height). A row of filter chips above it: All, Seating, Tables, Storage, Lighting (from the categories). Choosing a card loads that model from the internet (show a progress state on the card, and \"Could not load\" if it fails) and puts it on the stand, replacing what was there. My room: the pieces I have placed, each with its name and size and a remove button, the number of pieces, and the floor area they take together in square metres. \"Place in my room\" places the current piece where I point, on the real floor in AR, at its real size, and it stays there; I can place several pieces, drag any of them later, and turn the selected one with the slider. In VR and on a flat screen they are placed on the floor of the photographed room. Every model must appear at its true size: check the loaded model's bounding box against the catalogue's dimensions and correct the scale when a file is not in metres (some are ten times too large). Stand each piece on its base. While the list is loading show skeleton cards; with no connection show \"No connection\" and the lounge chair that ships with the app, so the app is never empty. Nothing may throw when the network is away. Keep the Device tab as it is. Title of the panel: \"Furniture catalogue\".","u":"/blog/a-furniture-catalogue-loaded-from-the-internet#the-request-word-for-word"},{"k":"blog","t":"A furniture catalogue loaded from the internet, placed in your room","h":"Notes and credits","x":"Models, thumbnails and sizes are loaded at run time from Poly Haven's public catalogue (CC0). The picture is a capture of the project running in Graspable's emulated headset.","u":"/blog/a-furniture-catalogue-loaded-from-the-internet#notes-and-credits"},{"k":"blog","t":"A furniture catalogue loaded from the internet, placed in your room","h":"Make your own","x":"Graspable is a desktop app. Describe the scene you want, and the agent builds it as a real project on your computer and checks that it runs. Download Graspable, and start from the Cross-platform XR template. More examples are in Made in Graspable.","u":"/blog/a-furniture-catalogue-loaded-from-the-internet#make-your-own"},{"k":"blog","t":"A penthouse sales suite in Unreal Engine, built from a description","x":"The living room of an apartment that is not built yet, for buyers to walk through in a headset. It was made as an Unreal Engine 5 project in two runs, and the first request is here word for word. This is the living room of an apartment that is not built yet, for buyers to walk through in a headset: low sun across an oak floor, a wall of glass onto a terrace, and a lever that changes day to evening","s":"The living room of an apartment that is not built yet, for buyers to walk through in a headset. It was made as an Unreal Engine 5 project in two runs, and the first request is here word for word. This is the living room of an apartment that is not built yet, for buyers to walk through in a headset: low sun across an oak floor, a wall of glass onto a terrace, and a lever that changes day to evening. It is an Unreal Engine project, so it does not run in a browser. Its app page shows the picture of it.","u":"/blog/a-penthouse-sales-suite-in-unreal-engine"},{"k":"blog","t":"A penthouse sales suite in Unreal Engine, built from a description","h":"What it shows","x":"- Unreal Engine 5 with Lumen. The picture is Unreal's desktop renderer. - Scanned furniture at real size. Furniture, surfaces and plants are scans. - A Blueprint written by the agent. - Inside a headset's budget. The request ends by asking what the room costs to draw, and the agent reported 107,476 triangles and 45 meshes, inside its budget for a standalone headset.","u":"/blog/a-penthouse-sales-suite-in-unreal-engine#what-it-shows"},{"k":"blog","t":"A penthouse sales suite in Unreal Engine, built from a description","h":"How it was made","x":"The project started from the Unreal Engine XR template: an Unreal Engine 5 project for OpenXR headsets, with scanned furniture and props at real size, daylight that bounces, and Blueprints the agent writes. It took two runs. The first request is below. The second request asked for a brighter day scene and a furnished terrace. The AI was Claude Code.","u":"/blog/a-penthouse-sales-suite-in-unreal-engine#how-it-was-made"},{"k":"blog","t":"A penthouse sales suite in Unreal Engine, built from a description","h":"The request, word for word","x":"This is the text that was sent to the agent to start the project. Nothing was edited for this article. We are selling the penthouse of our new development before it is built, and buyers will walk through it in a headset in our sales suite. Build its living room. The room is about 9 by 7 metres with a 3 metre ceiling. One long wall is glass, floor to ceiling in slim dark frames, onto a terrace with","s":"This is the text that was sent to the agent to start the project. Nothing was edited for this article. We are selling the penthouse of our new development before it is built, and buyers will walk through it in a headset in our sales suite. Build its living room. The room is about 9 by 7 metres with a 3 metre ceiling. One long wall is glass, floor to ceiling in slim dark frames, onto a terrace with a parapet. It is late afternoon and the sun comes in low across the floor. Furnish it the way an interior designer would for the brochure: a sofa group around a coffee table, an armchair, a sideboard with a few objects on it, plants, a ceiling lamp over the table. Oak floor, light plaster walls. Keep it calm and expensive: few things, well placed, nothing that looks like a toy. By the entrance, where I start, put a small wall panel with a lever. Lever up is the day scene. Lever down is the evening scene: the sun is gone, the lamps are on and the room is warm. A sign on the wall names the apartment: Residence 14B. It has to run on a standalone headset, so keep it light, and tell me what it costs to draw.","u":"/blog/a-penthouse-sales-suite-in-unreal-engine#the-request-word-for-word"},{"k":"blog","t":"A penthouse sales suite in Unreal Engine, built from a description","h":"Notes and credits","x":"The picture is a capture of the game running on a flat screen. It is Unreal's desktop renderer; a standalone headset uses the mobile renderer, without bounced light. No frame rate was measured, and it has not been tried on a headset. Furniture, surfaces and plants are scans from Poly Haven (CC0).","u":"/blog/a-penthouse-sales-suite-in-unreal-engine#notes-and-credits"},{"k":"blog","t":"A penthouse sales suite in Unreal Engine, built from a description","h":"Make your own","x":"Graspable is a desktop app. Describe the scene you want, and the agent builds it as a real project on your computer and checks that it runs. For an Unreal project you need Unreal Engine 5 on your computer. Download Graspable, and start from the Unreal Engine XR template. More examples are in Made in Graspable.","u":"/blog/a-penthouse-sales-suite-in-unreal-engine#make-your-own"},{"k":"blog","t":"A shelving configurator for Apple Vision Pro, from one request","x":"Choose width, height, finish and extras in a window, read the parts and the price, and see the shelf at real size in your room. A native app built in one run; the request is here word for word. This is a product configurator for a modular shelf. Choose width, height, finish and extras in a window, read the parts and the price, and see the shelf at real size in your room. It is a native app for App","s":"Choose width, height, finish and extras in a window, read the parts and the price, and see the shelf at real size in your room. A native app built in one run; the request is here word for word. This is a product configurator for a modular shelf. Choose width, height, finish and extras in a window, read the parts and the price, and see the shelf at real size in your room. It is a native app for Apple Vision Pro, so it does not run in a browser. Its app page shows the picture of it.","u":"/blog/a-shelving-configurator-for-apple-vision-pro"},{"k":"blog","t":"A shelving configurator for Apple Vision Pro, from one request","h":"What it shows","x":"- A product configurator. You choose width, height, finish and extras, and read the parts and the price. - Window and space. You choose in a window and see the result standing in the room. - Real materials. The oak is a photographed wood. The picture at the top is from the visionOS Simulator's living room. The window is titled Modu shelf: width, height, finish swatches and options on the left, a p","s":"- A product configurator. You choose width, height, finish and extras, and read the parts and the price. - Window and space. You choose in a window and see the result standing in the room. - Real materials. The oak is a photographed wood. The picture at the top is from the visionOS Simulator's living room. The window is titled Modu shelf: width, height, finish swatches and options on the left, a parts list and a total of 366 euros on the right. Beside the window stands an oak shelf with four compartments, doors with steel handles and a plinth, with a price tag and labels reading 155 cm and 80 cm.","u":"/blog/a-shelving-configurator-for-apple-vision-pro#what-it-shows"},{"k":"blog","t":"A shelving configurator for Apple Vision Pro, from one request","h":"How it was made","x":"The project started from the Apple Vision Pro template: a native app with a window of controls and an immersive space around the person, built and run in the visionOS Simulator by your own Xcode. It took one run. The AI was Claude Code.","u":"/blog/a-shelving-configurator-for-apple-vision-pro#how-it-was-made"},{"k":"blog","t":"A shelving configurator for Apple Vision Pro, from one request","h":"The request, word for word","x":"This is the text that was sent to the agent to start the project. Nothing was edited for this article. Turn this into a product configurator for modular shelving, the kind of app a furniture or retail company would commission for Apple Vision Pro: choose in a window, then see the result at real size in your own room. The product is a modular shelf built from RealityKit boxes: side panels, shelves ","s":"This is the text that was sent to the agent to start the project. Nothing was edited for this article. Turn this into a product configurator for modular shelving, the kind of app a furniture or retail company would commission for Apple Vision Pro: choose in a window, then see the result at real size in your own room. The product is a modular shelf built from RealityKit boxes: side panels, shelves and an optional back panel and optional doors on the lowest compartment. Real dimensions in metres. The window (make it wider, about 900 by 560 points) has two halves. Left: the options. - Width: 60, 80 or 120 cm (segmented). - Height: 3, 4 or 5 shelves (each compartment is 35 cm high). - Finish: Oak, Walnut, White, Graphite (colour swatches as round buttons; pick colours that look like those finishes). - Back panel: on or off. Doors on the bottom compartment: on or off. Right: a summary card with the name (\"Modu shelf\"), the overall size in cm (width x height x depth, depth is 32 cm), the list of parts with counts (side panels, shelves, back panel, doors), and the price, which changes with the options: make up a simple, sensible price list and show the total large. Below it a prominent button \"View in my room\" that opens the space, and becomes \"Close the room view\" while it is open. The space: the shelf stands on the floor 2.8 m in front of me and 1.1 m to the right of the window, turned a little toward me, so the whole shelf from the floor to its top is in view beside the window. It is built from the options and rebuilt at once when I change any of them, and can be dragged along the floor by pinching. Next to it float two dimension labels (width and height) with clear lines, and a price tag above it, all large enough to read from where I stand, on dark rounded backgrounds. Make it look like real furniture. Use the kit: Oak and Walnut are the photographed woods (Kit.material and Kit.box, with the grain running along each board), White is a smooth lacquer and Graphite a dark matt lacquer (Kit.plain). Boards are 18 mm thick with slightly rounded edges; shelves sit between the side panels; the back panel is thin; the doors have small brushed-steel handles and a narrow gap between them; the shelf stands on a low plinth. Keep the model in one observable class, with the price list and the part counts computed there, so the window and the space always agree. Keep the launch behaviour that opens the space when GRASPABLE OPEN SPACE is set. Start with: 80 cm, 4 shelves, Oak, back panel on, doors on.","u":"/blog/a-shelving-configurator-for-apple-vision-pro#the-request-word-for-word"},{"k":"blog","t":"A shelving configurator for Apple Vision Pro, from one request","h":"Notes and credits","x":"Prices and the product are made up. The wood is a photographed oak from Poly Haven (CC0). The picture is a capture of the app in the visionOS Simulator.","u":"/blog/a-shelving-configurator-for-apple-vision-pro#notes-and-credits"},{"k":"blog","t":"A shelving configurator for Apple Vision Pro, from one request","h":"Make your own","x":"Graspable is a desktop app. Describe the scene you want, and the agent builds it as a real project on your computer and checks that it runs. For a Vision Pro app you need a Mac with Xcode. Download Graspable, and start from the Apple Vision Pro template. More examples are in Made in Graspable.","u":"/blog/a-shelving-configurator-for-apple-vision-pro#make-your-own"},{"k":"blog","t":"How the National Theatre in Prague became a place to walk through together","x":"An open 3D model of the Národní divadlo by Lukáš Eršil was turned into a multi-user WebXR app in seven requests. Here is what it does, and every request word for word. Lukáš Eršil made an open 3D model of the National Theatre in Prague, the Národní divadlo. We found it through his post on X, and Michal Takáč thought it would be fun to test how far Graspable can take it. The result is a place for s","s":"An open 3D model of the Národní divadlo by Lukáš Eršil was turned into a multi-user WebXR app in seven requests. Here is what it does, and every request word for word. Lukáš Eršil made an open 3D model of the National Theatre in Prague, the Národní divadlo. We found it through his post on X, and Michal Takáč thought it would be fun to test how far Graspable can take it. The result is a place for several people at once. You walk the street, the foyer and the auditorium in a headset or a browser, dress for 1883, talk, and wind the building back to its foundations. Open the theatre. From the same page you can continue from it in Graspable. A room holds sixteen people. The National Theatre in Prague, walked through together.","u":"/blog/how-the-national-theatre-in-prague-was-built"},{"k":"blog","t":"How the National Theatre in Prague became a place to walk through together","h":"What it shows","x":"- Several people in one place. Everyone in a room walks the same theatre and sees the others as figures in the clothes of the time. - Hands and controllers. In a headset you can use either. - A menu on your hand. In the headset the menu is on your hand. - Chat in and out of the headset. People write to each other whether they came in a headset or in a browser. The longer film. How it was filmed is","s":"- Several people in one place. Everyone in a room walks the same theatre and sees the others as figures in the clothes of the time. - Hands and controllers. In a headset you can use either. - A menu on your hand. In the headset the menu is on your hand. - Chat in and out of the headset. People write to each other whether they came in a headset or in a browser. The longer film. How it was filmed is described below.","u":"/blog/how-the-national-theatre-in-prague-was-built#what-it-shows"},{"k":"blog","t":"How the National Theatre in Prague became a place to walk through together","h":"How the film was made","x":"The scenes with a crowd were filmed in a browser, and their fifteen visitors are simulated: scripted figures connected to a real room through the same relay people use, with invented names and lines. The scenes marked so in the film were recorded by one person in a Meta Quest 3.","u":"/blog/how-the-national-theatre-in-prague-was-built#how-the-film-was-made"},{"k":"blog","t":"How the National Theatre in Prague became a place to walk through together","h":"Whose model it is","x":"The model of the theatre is by Lukáš Eršil (github.com/lukasersil/narodni-divadlo-3d, MIT licence). His geometry, data and texts are used as they are. His sources and their licences (RÚIAN, IPR Praha, OpenStreetMap, photographs from Wikimedia Commons under CC BY-SA, the historic plans) are listed in the app under About the model. Walking, the shared state, the figures and their wardrobe, the menu ","s":"The model of the theatre is by Lukáš Eršil (github.com/lukasersil/narodni-divadlo-3d, MIT licence). His geometry, data and texts are used as they are. His sources and their licences (RÚIAN, IPR Praha, OpenStreetMap, photographs from Wikimedia Commons under CC BY-SA, the historic plans) are listed in the app under About the model. Walking, the shared state, the figures and their wardrobe, the menu in the headset, chat and teleporting were made in Graspable.","u":"/blog/how-the-national-theatre-in-prague-was-built#whose-model-it-is"},{"k":"blog","t":"How the National Theatre in Prague became a place to walk through together","h":"How it was made","x":"The project started from the Multi-user template: a room several people open at once, where everyone sees each other. It took seven runs: a first request and six that followed. They were sent one after another, each after looking at what the last one built. The AI was Claude Code with Claude Opus 5.5. It was tried on a Meta Quest 3 after the sixth request. Teleporting came with the seventh and has","s":"The project started from the Multi-user template: a room several people open at once, where everyone sees each other. It took seven runs: a first request and six that followed. They were sent one after another, each after looking at what the last one built. The AI was Claude Code with Claude Opus 5.5. It was tried on a Meta Quest 3 after the sixth request. Teleporting came with the seventh and has been checked in the emulated headset only.","u":"/blog/how-the-national-theatre-in-prague-was-built#how-it-was-made"},{"k":"blog","t":"How the National Theatre in Prague became a place to walk through together","h":"The seven requests, word for word","x":"Nothing below was edited for this article. The requests are long, and they are here in full, in the order they were sent. Request 1: the building, walking, and keeping everyone in step I found an incredible open project: an interactive 3D model of the National Theatre in Prague (Národní divadlo) by Lukáš Eršil. Its whole source is in reference/narodni-divadlo-3d (MIT licence; read its README.md fi","s":"Nothing below was edited for this article. The requests are long, and they are here in full, in the order they were sent. Request 1: the building, walking, and keeping everyone in step I found an incredible open project: an interactive 3D model of the National Theatre in Prague (Národní divadlo) by Lukáš Eršil. Its whole source is in reference/narodni-divadlo-3d (MIT licence; read its README.md first, then index.html and src/). It is a single page you orbit around with the mouse. I want to take it further: make it a place several people walk through together, on a screen and in a headset. This first step is the building itself, walking, and keeping everyone in step. Menus in the headset, avatars and chat come in later requests, so leave room for them but do not build them now. 1. Bring the whole model into this project - Everything Lukáš's model has must be here and work the same: the seven construction stages on the build clock (0 to 7) with their animations, Play through the whole construction, Exterior and X-ray with the ten layers (each on or off, all, none), the long section and the horizontal section with their two sliders, day and night, the fire of 1881 with the collapse and \"Rebuild 1883\", the eight viewpoints, the hotspots with their cards and pictures, Czech and English, and \"About the model\" with every source and licence he lists. - Reuse his code for the geometry, the registry and the data as it is wherever you can; do not redraw the theatre by hand. His page uses three.js r186 with the WebGPU renderer and node materials. This project uses React Three Fiber and has to run in WebXR on a standalone headset, where the WebGL renderer is the safe choice: keep one renderer for screen and headset, and port the materials and effects (night lighting, the X-ray ghost look, the section planes with the dark cut walls, the fire, the sky) to it so they look as close to his as you can get. Update three.js if his code needs a newer version, and check that the multi-user kit and the XR emulator still work afterwards. - Copy the pictures and textures the model needs from reference/narodni-divadlo-3d/assets into public/, except avatar.png: that is his portrait and signature, do not use it. - Remove the lounge, the table objects and the board of the starting template. The place is now the theatre and the streets around it. 2. Walking - Real size, 1:1. People start on Národní street in front of the main façade, facing it, a few steps apart from each other. - On a computer: W A S D to walk, arrow keys left and right to turn, arrow keys up and down to look up and down, Shift to walk faster, and dragging with the mouse to look around. No orbit camera. - In a headset: left thumbstick to walk in the direction you look, right thumbstick to turn in steps. - Physics: nobody walks through walls, columns, railings or closed parts of the building, and you stand on what is under you (street, steps, floors, the stage, the tiers). Use the physics engine that is already in the project, with colliders made from the model's own geometry, kept simple enough for a headset. Colliders must follow the model: a wall that is not built yet at the current stage, or is cut away by the section, or has collapsed in the fire, does not stop anyone. - One setting, \"Walk through walls\", off at the start. When a person turns it on, only they pass through everything and can also move up and down (E and Q on a computer, right thumbstick up and down in a headset), so they can reach the dome or the cellars. - The eight viewpoints become places you are taken to: you arrive standing at that spot looking the way his camera looks. In a headset this is an instant move with a short fade, never a camera flight. Viewpoints that are in mid-air (above the roofs, from the river) put the person there with \"Walk through walls\" on, and say so. - Add \"Back to the street\" which returns you to the start. 3. Together - Everyone in the room sees the same building: the build clock and Play, Exterior or X-ray and the layers, the section and its sliders, day or night, the fire and the rebuild are shared. When one person changes them, everyone sees it. Language, viewpoints, open cards, \"Walk through walls\" and where you stand are each person's own. - A person who joins later sees the current state at once. - Keep the template's avatars for now so people can see each other; they must be drawn where each person really stands in the theatre. - It must work alone, and without a connection to the relay. 4. Controls on a computer for now - On a flat screen, keep Lukáš's interface as an HTML layer over the canvas: his toolbar, side panel with viewpoints, section sliders and layers, the stage card, the timeline with Play, the fire button and card, hotspot markers and cards, About. Keep his visual style (black panels, white type, one yellow accent, Bebas Neue and DM Sans), but replace his masthead with one that reads \"Národní divadlo\" and under it \"Model by Lukáš Eršil. Multi-user WebXR version made in Graspable.\" Add the \"Walk through walls\" switch and \"Back to the street\" to the toolbar. This HTML layer is wanted here, on purpose; it is hidden while a headset session runs. - Keep the \"Enter VR\" button. 5. It has to run well in a headset - His page draws about 830 meshes in about 1,000 draw calls and 1.4 million triangles, with a 4096 shadow map and a bloom pass. That is too much for a standalone headset. In a headset session aim for under 300 draw calls and under 700,000 triangles in view at 72 frames a second on a Quest 3: merge what never moves apart once a stage is finished, draw the far surroundings cheaply, do not draw the inside while you are outside and the other way round unless X-ray or a section needs it, no real-time shadow map in the headset (a baked or simple contact shadow is enough), no post-processing pass in the headset. - On a computer keep his quality: shadows and the glow at night. - The construction animation, X-ray, section and fire must still work after merging: it is fine to switch between a merged finished state and the separate animated pieces while something animates. - Show a small readout I can switch on with the P key: frames per second, draw calls and triangles. 6. Credit - \"About the model\" keeps all of Lukáš's sources and licences word for word (RÚIAN, IPR Praha, OpenStreetMap, the Wikimedia photographs with their CC BY-SA authors, the historic plans), and adds at the top: \"The model is by Lukáš Eršil (github.com/lukasersil/narodni-divadlo-3d, MIT). We found it through his post on X (x.com/lukasersil/status/2107068853154992567), and Michal Takáč thought it would be fun to test how far Graspable can take it: this multi-user WebXR version was made in Graspable.\" in English, and the same in Czech. - Keep his LICENSE file in the project next to the code that comes from him, and a NOTICE.md at the root that lists what comes from his project and the licences of the textures and data. When you are done, check it yourself: the page loads without errors, you can walk from the street up to the façade and are stopped by it, each of the modes above works, and two browser windows in the same room see each other and the same building state. Tell me what you could not make work and what you could not check. Request 2: fixes after a first walk on a computer The theatre is here and the interiors look very close to Lukáš's. I went through it on a computer and found these. Fix them before we go on; reproduce each one first, in the same browser you check with. 1. The picture freezes - Open the page, choose the viewpoint \"Foyer\", then \"Národní Street\". The interface reacts (the readout changes, the hotspot markers move to where they would be on the street) but the 3D picture still shows the foyer. It stays frozen through \"From the river\", \"Dome\" and changes of stage, and only comes back when something like the fire or \"From the stage\" starts. - The same after the fire: go to \"From the river\", press \"Fire of 1881\", wait for the collapse, press \"Rebuild 1883\", wait eight seconds. The picture shows the finished building with black smoke standing over it and never changes again: choosing \"Load-bearing walls\" changes the readout to 172 draw calls but the picture still shows the finished building and the smoke, and \"Back to the street\" leaves the picture at the river. - I suspect the moment the glow pass switches off (leaving a room, the fire ending), but find the real cause. After the fix, the picture must always show the current state. Go through every change between glow and no glow: indoors to outdoors and back, night on and off, X-ray on and off, fire start, collapse, rebuild, and a stage change in the middle of each. - When the fire is over (rebuild finished, or any timeline action during the fire, as in his page) no flames or smoke are left. 2. Where people start - The start is too close to the façade and between two lamp posts: you see a lamp post and three doors, not the theatre. Start people on the far pavement of Národní, opposite the middle of the main façade, clear of lamp posts, trees and tram tracks, looking at the entrance with the head tilted up a little so the loggia and the trigas are in view. Several people start side by side along that pavement, not behind each other. - Trams must not run through people standing at the start. 3. Night - At night from the start a street lamp near the camera turns into a white blob that covers a third of the picture. Lamps should glow as in his page: bright globes with a soft halo, never a white-out. Check the night from the start, from the loggia and from the river. 4. One way into the headset - There are two buttons, \"Enter XR\" at the top and \"Enter VR\" in the masthead. Keep one, in the masthead, in the interface's style; it says \"Enter VR\" (or \"Enter AR\" is not offered: this is a place, not a room overlay). 5. Sections and X-ray in the headset - In the cut-away half of a section a person falls about 24 m. Nobody should fall because of a display mode: with a section on, the floors, stairs and ground stay solid everywhere, also where they are cut away and not drawn; only the walls that are cut away let people through. - X-ray and the section draw about 1.1 to 1.2 million triangles per eye in the headset. Bring both under the 700,000 target there: in X-ray and in a section do not draw the far surroundings, the trams and the trees in the headset, simplify what the ghost shell draws (it is a transparent hint, it does not need every baluster and window frame), and say what the numbers are afterwards. On a computer keep everything. 6. Phones and tablets - Many people will open this from a link on a phone. A touch screen needs walking too: a thumb stick in the lower left corner to walk, dragging anywhere else to look around, two buttons for up and down when \"Walk through walls\" is on. Keep Lukáš's phone layout for the interface (his style sheet already has one) and make sure the thumb stick and the timeline do not cover each other. On a phone use the headset's lighter drawing (merged pieces, no shadow map, no glow) so it runs. 7. Small things - Starting the page must not show the section as switched on unless it is. - \"Play\" on a computer: since nobody is flown around any more, a person standing at the façade sees little of the construction. When a person presses Play while standing on the street, offer one line \"Watch from the river\" that takes them to that viewpoint; do not move anyone without asking. Then check again as before, plus: every item above, two windows in one room, and the emulated headset. Give me the readout numbers on the street, in the auditorium, in X-ray and in a section, for the computer and for the emulated headset. Request 3: the menu inside the headset Now the menu inside the headset. Everything the HTML interface does on a computer must be possible from within XR, with the same wording and the same look. Build it with @react-three/uikit (add the package): real 2D panels in the 3D scene, in Lukáš's style (black panels, white type, one yellow accent for what is active, square corners; Bebas Neue for headings and DM Sans for text, loaded as uikit fonts and shipped with the project so it works offline). 1. Where it is - The menu is attached to the left hand: to the left controller, or with hand tracking to the left wrist, tilted towards the face like a watch you look at. It is operated with the right hand: the controller's ray and trigger, or with bare hands by touching it with the index finger (and a pinch from a distance). - It opens and closes with the left controller's menu or Y button, and with bare hands by turning the left palm towards the face. Closed, nothing is in view. - One button \"Pin\" leaves the open menu standing in the room where it is, so several people can gather around their own; \"Unpin\" brings it back to the hand. - Left-handed people can switch sides in the menu. - Readable: the smallest text is at least 1.5% of its distance from the eyes in height (at 45 cm that is about 7 mm letters). Check sizes by measuring, not by looking at an emulator screenshot. 2. What is in it (tabs across the top; the same actions as on the computer) - Timeline: the eight stages as buttons with their years and names, the current one marked; a slider for the build clock; Play and Pause; the stage card with its title and text below. - View: Exterior / X-ray; the ten layers with their colour dots, each on or off, All, None (only in X-ray); Section on or off with the two sliders and their values (long section in metres from the axis, horizontal section in metres above sea level); Day / Night. - Fire: \"Fire of 1881\" with its card (title, text, the engraving and its credit) and \"Rebuild 1883\". - Places: the eight viewpoints as buttons, \"Back to the street\", and the \"Walk through walls\" switch. - About: language CZ / EN, the About text with all sources, licences and the credit to Lukáš Eršil, scrollable. - Sliders must be easy to drag with a ray and with a fingertip, and show their value while dragged. 3. Hotspots in the scene - In a headset the numbered hotspots are small round markers standing in the building where they belong, always turned to the viewer, the same ones the current stage or X-ray layer shows on the computer. Pointing and pressing the trigger, or touching one, opens its card as a panel beside the marker facing the person: year or layer, title, text, the picture with its credit, and a close button. Only one card is open at a time. 4. The same for everyone and for each - Shared things changed in the menu (clock, Play, X-ray and layers, section, night, fire) change for everybody, exactly as on the computer. Each person's menu shows the current shared state, also when someone else changed it. 5. Do not lose frames - The menu and the cards must not cost more than a few draw calls when open and nothing when closed. Text that does not change is not redrawn every frame. Check it in the emulated headset: open the menu, press through every tab, drag both section sliders, start the fire and rebuild, open a hotspot card, switch the language. Tell me what you could not check without a real headset. Request 4: the people and their wardrobe Now the people. Replace the template's avatars with full figures dressed as people were in Prague while the theatre was being built, 1868 to 1883, and let everyone choose who they are. 1. The figures - A whole person at real size, about 1.6 to 1.8 m: head, body, arms and legs, built from simple shapes in code (no downloaded models), stylised like the statues in Lukáš's model, with clean silhouettes and cloth colours of the time. Each figure is one or two draw calls and a few thousand triangles at most, so twenty people in the room cost little. - The head follows where the person looks. In a headset the hands follow the controllers or tracked hands and the arms follow the hands. On a computer the arms hang and swing while walking. Legs step while the person moves and stand still when they stand. The body turns smoothly towards the walking direction. - The name plate from the multi-user kit stays above the head. - You do not see your own head from the inside; looking down you see your own body and hands. 2. The wardrobe: at least eight outfits that are right for the years 1868 to 1883, each with a name in Czech and English - A gentleman in a black frock coat, waistcoat, cravat and top hat. - A burgher in a čamara, the braided Czech patriotic coat, with a soft round hat. - A lady in a bustle dress with a fitted bodice, gloves and a small hat with a ribbon. - A young woman in a day dress with an apron-front overskirt and a bonnet. - A stonemason from the building site: shirt with rolled sleeves, waistcoat, leather apron, flat cap. - A carpenter: linen shirt, braces, rolled trousers, a soft cap. - A Sokol member in the red shirt, the jacket worn over one shoulder and the round cap with a feather. - A painter of the \"National Theatre generation\": velvet jacket, loose bow tie, wide beret, with a brush in the breast pocket. - For every outfit the person can pick one of four cloth colours that suit it, and one of four skin tones. Get the cuts right: frock coats to the knee, the bustle at the back, 1870s hats. 3. Choosing, in a headset - A new tab \"Wardrobe\" in the hand menu: the outfits as a list with small pictures, the colours and skin tones as swatches, and a field for the name shown on the plate. - \"Mirror\" in that tab stands a tall mirror in a gilded frame about 1.5 m in front of the person, facing them: a real mirror that shows them in their outfit, moving as they move, and whoever stands beside them. It stays until closed, and only the person who opened it sees it. Draw the mirror at a modest resolution and only while it is open, so the headset keeps its frame rate. - There is also one fixed mirror for everyone in the foyer, on the wall between the doors, lit, with a plate that reads \"Šatna · Wardrobe\". 4. Choosing, on a computer - A \"Wardrobe\" button in the toolbar opens a panel in the same style: the same list, swatches and name field, and a live preview of the figure turning slowly, drawn in a small canvas of its own. Pressing M shows your own figure from the front for a moment in the 3D view (the camera steps back and returns). 5. For everyone - The choice travels with the person: others see the outfit, colour, skin tone and name at once, and a person who joins later sees everyone as they are. Keep what is sent small (a few numbers and the name), through presence, not through shared state. - The choice is remembered on this device for the next visit. - A newcomer gets a random outfit, so a room full of people looks like a street of 1880 and not like copies. Check with two browser windows: each sees the other's figure walk, turn its head, and change outfit; the mirror shows your own figure; the readout (P) with five figures in view stays inside the budget from before. Tell me what you could not check. Request 5: chat, from a computer and from the headset Last part: people should be able to write to each other, from a computer and from inside the headset. 1. The chat - Short text messages for everyone in the room. Each shows the sender's name in their colour, the text, and the time. The last 50 are kept for the room, so a person who joins sees what was said before; older ones are dropped. Messages are at most 240 characters, plain text only: nothing a person types is ever treated as markup or a link. - When a message arrives, a speech bubble with its text stands above the sender's head for six seconds, turned to the viewer, readable from ten metres, and then fades. So you see who is speaking in the street. - People can also be seen typing: three dots above the head while they write. 2. On a computer - A chat panel in the lower left corner in the same style as the rest: the message list, and a field you type in. Enter sends; T or Enter puts the cursor into the field; Escape leaves it. While the cursor is in the field, W A S D and the other keys type letters and do not move the person. - The panel can be folded to a small tab that shows the number of unread messages. 3. In a headset - A new tab \"Chat\" in the hand menu: the message list, scrollable, and below it a keyboard built in the same uikit style, large enough to type on with two controller rays or with two index fingers: Czech and English letters (a key switches between the QWERTZ layout with the Czech accented letters and QWERTY), digits, space, backspace, shift, and Send. Keys answer with a short pulse on the controller and a quiet click. - Above the keyboard a row of ready phrases in the menu's language that send with one press: \"Hello!\", \"Look at this\", \"Follow me\", \"Meet me in the foyer\", \"To the stage!\", \"Beautiful, isn't it?\", \"Thank you\". - Where the browser offers speech recognition, a microphone key writes what the person says into the field for them to check and send; hide the key where it is not offered. Nothing is recorded or sent anywhere by us. - With the menu closed, a new message shows for a few seconds as a small line near the left hand, with the sender's name. 4. Good manners - A person can mute another from the list of people in the room (a new small list in the Chat tab and in the computer panel: name, outfit, and \"Mute\"). Muted people's messages and bubbles are not shown to the person who muted them. - At most one message a second per person. 5. Everything together - Go through the whole thing once more with two browser windows, one of them in the emulated headset: walking and collisions, the timeline and Play, X-ray and layers, both sections, night, the fire and the rebuild, every viewpoint, hotspot cards, language, the wardrobe and mirrors, and chat in both directions. Fix what does not hold. Then give me the readout numbers (frames per second, draw calls, triangles) on the street, in the auditorium and in X-ray, on the computer and in the emulated headset, and a list of what still needs a real headset to be sure. Request 6: a last pass on how it looks It all works together. A last pass on how it looks before I show it to people. 1. People at night - With Night on, the figures are almost black: you see the name plate and the speech bubble but not the person or the outfit. People standing near the street lamps and the lit façade should be lit by them, warm and soft, so a top hat, an apron or a red Sokol shirt can still be told apart at ten metres. Do it without real lights per lamp (the headset cannot afford them): a night tint and a little light of their own on the figures is enough. Check from the start and inside the foyer and auditorium, by day and by night. 2. My own arms in the headset - Looking down in the headset, my arms are two long straight tubes from the shoulders to the controllers, like stilts. Arms have an elbow: bend them (upper arm and forearm of fixed, natural lengths, the elbow down and slightly out) so the hand reaches the controller the way an arm does. When the controller is further away than the arm is long, the arm stretches out straight and stops there; it never gets longer. The same for how others see a headset wearer. 3. The menu's tabs - With seven tabs the labels are too small and crowded. Put the tabs in two rows, or give each a simple pictogram above a short label, so each is as easy to hit and read as the buttons below them. Keep the smallest text at the size rule from before. - With the menu on the hand, the Chat keyboard hangs so low that its lower rows are out of view. When the Chat tab is open and the menu is not pinned, place the list and the keys so both are in view at a natural arm position, or pin the menu automatically in front of the person while they type and put it back afterwards. Say which you chose. 4. Small things - The page asks for favicon.ico and gets a 404: add a small icon (a yellow square with the dome's outline in black is enough). - On a computer at the start, other people who stand closer to the theatre are cut off by the timeline at the bottom of the screen. Start with the head tilted up a little less, so both the loggia and the people on the street are in view. Check each in the browser and the emulated headset, re-record the headset tests that change, and give me the readout numbers on the street by day and by night with five people in view. Request 7: teleporting I tried it in a Meta Quest 3 and it works very well. One thing is missing before I show it to people: teleporting. Walking with the thumbstick makes some people queasy, so they need to be able to point at a spot and be there. 1. Teleport with controllers - Push the right thumbstick forward and hold: an arc leaves the right controller and lands on what is under it, with a ring where you would stand. Release the stick and you are there, behind the same short fade as the viewpoints. While the arc is out, turning the stick left or right before releasing turns the ring's arrow, and you arrive facing that way. - The right stick keeps its other jobs: left and right alone still turn in steps, and with \"Walk through walls\" on, up and down still move you up and down (so with it on, teleport is on the left controller's trigger while pointing, or say what you chose). - Left-handed people who switched the menu's side get teleport on the other hand. 2. Teleport with bare hands - Point with the right hand, palm up, index finger out: the same arc and ring appear. Pinch thumb and index finger to go there. It must not fire while the hand is busy with the menu, a slider, the keyboard or a hotspot. 3. Where you may land - Only on what a person can stand on: the street, pavements, steps, floors, the stage, the tiers and balconies, the loggia. Never on a wall, a roof slope, a statue, the river, a tram or another person. A spot you may not land on shows the ring in red and nothing happens on release. - The same rules as walking: what is not built yet at the current stage, or has collapsed in the fire, cannot be landed on; with a section on, floors stay floors. - At most 12 metres at a time, and never through a closed wall: the arc stops at the first thing it hits. Through an open arch or a door opening is fine. - With \"Walk through walls\" on, you may land anywhere the arc reaches, also in mid-air at its end. 4. For everyone else - Others see you disappear and arrive, not slide: your figure fades out for a moment where you were and in where you land. 5. On a computer and a phone - Double-click (or double-tap) on the ground does the same: you are taken there, facing the way you were looking. A place you may not land on does nothing. 6. Menu and help - The Places tab gets one line that says how to teleport with what is in the person's hands right now (controllers or bare hands), in Czech and English, and a switch \"Teleport only\" that turns off thumbstick walking for people who never want it. - The hint shown at the start on a computer mentions the double-click. Check it in the emulated headset with controllers and with hands: land on the street, on the loggia steps, inside the foyer through the open arches, on a tier of the auditorium; try a wall, the river and a roof and see the red ring; turn the arrow before landing. Check that a second window sees the arrival. Re-record the headset tests that change and add one for a teleport. Give me the readout with the arc showing, and tell me what still needs the real headset.","u":"/blog/how-the-national-theatre-in-prague-was-built#the-seven-requests-word-for-word"},{"k":"blog","t":"How the National Theatre in Prague became a place to walk through together","h":"Make your own","x":"Graspable is a desktop app. Describe the scene you want, and the agent builds it as a real project on your computer and checks that it runs. Download Graspable, and start from the Multi-user template if you want a place people share. More examples are in Made in Graspable.","u":"/blog/how-the-national-theatre-in-prague-was-built#make-your-own"},{"k":"blog","t":"Agents, wallets and cameras: staying in charge of your own","x":"An AI agent that reads the web, a wallet it can spend from, and a headset with cameras on your face each widen what can go wrong. A practical list of what to check, for people who use such tools and for people who build apps for them. Three things arrived in everyday software within about a year. Agents that act for you, wallets those agents can spend from, and headsets and glasses that see the ro","s":"An AI agent that reads the web, a wallet it can spend from, and a headset with cameras on your face each widen what can go wrong. A practical list of what to check, for people who use such tools and for people who build apps for them. Three things arrived in everyday software within about a year. Agents that act for you, wallets those agents can spend from, and headsets and glasses that see the room you are in. Each is useful alone. Together they mean that software you did not write can read what a stranger wrote, hold money, and look through a camera you are wearing. This post is a checklist. It names no product, and it applies to any tool of this kind, ours included.","u":"/blog/security-for-agents-wallets-and-headsets"},{"k":"blog","t":"Agents, wallets and cameras: staying in charge of your own","h":"Text an agent reads is not an instruction","x":"An agent reads files, web pages, messages, tool results and other people's code. Any of them can contain a sentence addressed to the agent. OWASP lists this, prompt injection, as the first risk for applications built on language models, and says plainly that no complete fix exists. What you can check in a tool: - It asks before it runs a command or reaches outside your project, and the question sh","s":"An agent reads files, web pages, messages, tool results and other people's code. Any of them can contain a sentence addressed to the agent. OWASP lists this, prompt injection, as the first risk for applications built on language models, and says plainly that no complete fix exists. What you can check in a tool: - It asks before it runs a command or reaches outside your project, and the question shows the real command. - A file that runs code, such as a build script, is treated like a command when the agent changes it. - What it remembers between sessions is something you can read, edit and delete. - Tools from a third party cannot declare themselves safe. You decide what runs without asking. - A project you got from someone else is treated as a stranger's text until you have read it. A habit that helps more than any setting: read the approval card. Most successful attacks on agents end with a person clicking yes.","u":"/blog/security-for-agents-wallets-and-headsets#text-an-agent-reads-is-not-an-instruction"},{"k":"blog","t":"Agents, wallets and cameras: staying in charge of your own","h":"A wallet for an agent is a spending account","x":"Give an agent a wallet the way you would give a teenager a prepaid card. The safest design keeps the key out of the model's reach entirely, so the model can ask for a payment and something else, with rules, makes it. Before you fund one: - Use a wallet made for the agent. Never your main wallet, and never a seed phrase you use elsewhere. - Put in only what you accept losing. - Set limits for one p","s":"Give an agent a wallet the way you would give a teenager a prepaid card. The safest design keeps the key out of the model's reach entirely, so the model can ask for a payment and something else, with rules, makes it. Before you fund one: - Use a wallet made for the agent. Never your main wallet, and never a seed phrase you use elsewhere. - Put in only what you accept losing. - Set limits for one purchase, one task and one day, and confirm they are enforced by the program and not by the model's good sense. - Prefer payments that authorise one exact transfer over approvals that let a contract spend up to a ceiling. EIP-3009 is an example of the first kind. - Ask to be consulted when the reason for a purchase comes from something the agent read. - Keep a record you can read later: what was bought, for which task, and who decided. Be wary of any setting that lets an agent spend with no questions. If you use one, the limits are your whole protection.","u":"/blog/security-for-agents-wallets-and-headsets#a-wallet-for-an-agent-is-a-spending-account"},{"k":"blog","t":"Agents, wallets and cameras: staying in charge of your own","h":"Keep your own keys and your own data","x":"Sovereignty here means something concrete. You can leave, and nobody can act as you. - Keys and sign-ins belong in your operating system's keychain or a hardware wallet. A plain file is not a place for a secret. - A service that takes payment from your wallet needs an address to pay and nothing else. It never needs your key, and it has no reason to hold funds for you. - Know which of your data lea","s":"Sovereignty here means something concrete. You can leave, and nobody can act as you. - Keys and sign-ins belong in your operating system's keychain or a hardware wallet. A plain file is not a place for a secret. - A service that takes payment from your wallet needs an address to pay and nothing else. It never needs your key, and it has no reason to hold funds for you. - Know which of your data leaves your computer. Prompts, files and screenshots are different from counts and error reports, and a tool should say which it sends and let you switch it off. - Prefer tools that work with your own AI account as well as theirs, and that keep your project as ordinary files you can open without them. - Give every outside program its own access token with the least it needs, and revoke tokens you no longer use.","u":"/blog/security-for-agents-wallets-and-headsets#keep-your-own-keys-and-your-own-data"},{"k":"blog","t":"Agents, wallets and cameras: staying in charge of your own","h":"Cameras on your face","x":"A headset or a pair of glasses with cameras sees your home, your screen, your documents and the people near you. On the web, the WebXR specification requires a user's consent before an immersive session starts and treats room and pose data as sensitive. That consent is the control you have, so use it deliberately. - Grant camera, microphone and room access per app and per session. Refuse an app th","s":"A headset or a pair of glasses with cameras sees your home, your screen, your documents and the people near you. On the web, the WebXR specification requires a user's consent before an immersive session starts and treats room and pose data as sensitive. That consent is the control you have, so use it deliberately. - Grant camera, microphone and room access per app and per session. Refuse an app that asks for more than it uses. - Open apps from strangers in a browser first, where permissions are visible, before installing anything. - Assume a recording light can be missed. Tell people around you when you are capturing. - Cover or take off the device near screens with secrets on them, such as a password manager or a wallet. - An assistant that sees through your camera can read text in the room. Treat what it sees as text a stranger wrote, with the same caution as a web page.","u":"/blog/security-for-agents-wallets-and-headsets#cameras-on-your-face"},{"k":"blog","t":"Agents, wallets and cameras: staying in charge of your own","h":"If you publish apps for others","x":"People who open your app are trusting you with their device. - Ask for the fewest permissions, at the moment they are needed, and say why. - Never put an API key, a private key or a sign-in of your own inside an app that others load. Anything in the page can be read by its visitors. - If your app reaches one of your own accounts, restrict it to the one sheet, folder or channel it needs, and to rea","s":"People who open your app are trusting you with their device. - Ask for the fewest permissions, at the moment they are needed, and say why. - Never put an API key, a private key or a sign-in of your own inside an app that others load. Anything in the page can be read by its visitors. - If your app reaches one of your own accounts, restrict it to the one sheet, folder or channel it needs, and to reading where reading is enough. - Keep each app at its own web address, so one app cannot read another's storage. - Tell people what you collect. A scan of someone's living room is personal data.","u":"/blog/security-for-agents-wallets-and-headsets#if-you-publish-apps-for-others"},{"k":"blog","t":"Agents, wallets and cameras: staying in charge of your own","h":"A short list to keep","x":"Before you switch on a new agent feature, ask these: 1. Can I see and stop every action that leaves my project or spends money? 2. Is the money it can reach an amount I accept losing? 3. Do I hold the keys, and can I leave with my work? 4. Does each app and device see only what it needs?","u":"/blog/security-for-agents-wallets-and-headsets#a-short-list-to-keep"},{"k":"blog","t":"Agents, wallets and cameras: staying in charge of your own","h":"Sources","x":"- OWASP Top 10 for LLM applications - OWASP Top 10 for agentic applications - Model Context Protocol: security best practices - EIP-3009: transfer with authorization - x402, a protocol for payments over HTTP - WebXR Device API, W3C - MediaDevices.getUserMedia, MDN - Securing your wallet, bitcoin.org","u":"/blog/security-for-agents-wallets-and-headsets#sources"},{"k":"blog","t":"Profiles, follows, likes and collections","x":"Graspable has a community side now: a public page for what you have made, people to follow, apps to like and keep in collections, and a gallery of everything public. Here is what each part does. Publishing an app from Graspable gives it an address. That was the end of it: you sent the link to someone, and nobody else ever found the app. There is now a place for apps to be found and for the people ","s":"Graspable has a community side now: a public page for what you have made, people to follow, apps to like and keep in collections, and a gallery of everything public. Here is what each part does. Publishing an app from Graspable gives it an address. That was the end of it: you sent the link to someone, and nobody else ever found the app. There is now a place for apps to be found and for the people who make them to be followed.","u":"/blog/community-profiles-follows-collections"},{"k":"blog","t":"Profiles, follows, likes and collections","h":"Your page","x":"Choose a handle and you get a public page at graspable.dev/@you . It lists your public apps with their pictures, how often each was viewed and liked, your followers, your public collections and the month you joined. You fill it in on the website under Account → Profile : - a name and a few words about you - where you are - up to five links, each shown with the name you have on that service - the h","s":"Choose a handle and you get a public page at graspable.dev/@you . It lists your public apps with their pictures, how often each was viewed and liked, your followers, your public collections and the month you joined. You fill it in on the website under Account → Profile : - a name and a few words about you - where you are - up to five links, each shown with the name you have on that service - the headsets and tools you build with - up to three apps to show first - whether you are open to work A public profile, signed out, as anyone sees it. Your email address is never shown. The picture is the one from the GitHub or Google account you sign in with, and it stays off until you switch it on.","u":"/blog/community-profiles-follows-collections#your-page"},{"k":"blog","t":"Profiles, follows, likes and collections","h":"Follow, like, save, share","x":"Every public app has a page on graspable.dev with its pictures, a longer text from its maker, and buttons to open it, like it, save it and share it. The maker's card on that page has Follow . An app's page: open it, continue from it, like it, save it, share it. A like is one per account, and clicking again takes it back. Following someone puts what they publish next under People you follow in the ","s":"Every public app has a page on graspable.dev with its pictures, a longer text from its maker, and buttons to open it, like it, save it and share it. The maker's card on that page has Follow . An app's page: open it, continue from it, like it, save it, share it. A like is one per account, and clicking again takes it back. Following someone puts what they publish next under People you follow in the app's Community page. Saving puts an app into a collection, which is a named list with a page of its own that you can send to someone. Under Account → Collections you choose which of them anyone can see. All of it works on the website and in the app with the same account. Looking needs no account at all.","u":"/blog/community-profiles-follows-collections#follow-like-save-share"},{"k":"blog","t":"Profiles, follows, likes and collections","h":"The gallery","x":"graspable.dev/made lists every public app. You can search it by name, tag or maker, filter by category, and order it by trending, most viewed, most liked or newest. Trending is this week's views and likes, with a like counting as five views. The gallery on 7 October 2026, with the counts as they were. To be found there, give your app a category and up to five tags under Settings → Published apps .","s":"graspable.dev/made lists every public app. You can search it by name, tag or maker, filter by category, and order it by trending, most viewed, most liked or newest. Trending is this week's views and likes, with a like counting as five views. The gallery on 7 October 2026, with the counts as they were. To be found there, give your app a category and up to five tags under Settings → Published apps . The examples our own team made sit in the same gallery, on the same kind of page as everyone else's.","u":"/blog/community-profiles-follows-collections#the-gallery"},{"k":"blog","t":"Profiles, follows, likes and collections","h":"Made with Graspable","x":"A public app carries a small Made with Graspable badge that leads to its page. From there Continue from this app opens it as a new project on the computer of whoever clicks, ready for their next request. This is on unless you switch it off. Switching it off needs the Builder plan or higher, and the badge goes with it. A private app is in none of these lists. It cannot be liked, collected or contin","s":"A public app carries a small Made with Graspable badge that leads to its page. From there Continue from this app opens it as a new project on the computer of whoever clicks, ready for their next request. This is on unless you switch it off. Switching it off needs the Builder plan or higher, and the badge goes with it. A private app is in none of these lists. It cannot be liked, collected or continued from.","u":"/blog/community-profiles-follows-collections#made-with-graspable"},{"k":"blog","t":"Profiles, follows, likes and collections","h":"Counts are real","x":"A view is counted when an app is opened at its address. A like exists only because an account clicked it. Nothing in the gallery is seeded, and no account exists to make it look busier than it is. It is a small place today, and the numbers say so.","u":"/blog/community-profiles-follows-collections#counts-are-real"},{"k":"blog","t":"Profiles, follows, likes and collections","h":"Start","x":"Publish something, open Settings → Graspable account to choose a handle, and edit your profile. The details are in the documentation under Publish and share.","u":"/blog/community-profiles-follows-collections#start"},{"k":"blog","t":"An agent with a wallet and a budget","x":"Some requests need something that costs money. Graspable can let the agent pay for an outside service from a wallet of its own, inside limits you set, with every purchase and every refusal written down on your computer. An agent building your scene sometimes hits something that costs money. A web search that answers properly, a dataset, a model that does one thing well, an hour of compute. Until n","s":"Some requests need something that costs money. Graspable can let the agent pay for an outside service from a wallet of its own, inside limits you set, with every purchase and every refusal written down on your computer. An agent building your scene sometimes hits something that costs money. A web search that answers properly, a dataset, a model that does one thing well, an hour of compute. Until now it stopped and told you. With agent spending switched on, it can buy from services that are paid per call, inside limits you set. It is experimental and off until you switch it on. Agent spending in about ninety seconds.","u":"/blog/how-the-agent-pays-for-things"},{"k":"blog","t":"An agent with a wallet and a budget","h":"A wallet of its own","x":"Under Settings → Agent spending you make a hot wallet. Graspable creates it on your computer and keeps its key in your keychain. The wallet is yours and you can show its key at any time. It is separate from your personal wallets, and Graspable never asks for those. You send it USDC on the Base network. The wallet can spend only what you sent, so keep in it what you would keep in an app and no more","s":"Under Settings → Agent spending you make a hot wallet. Graspable creates it on your computer and keeps its key in your keychain. The wallet is yours and you can show its key at any time. It is separate from your personal wallets, and Graspable never asks for those. You send it USDC on the Base network. The wallet can spend only what you sent, so keep in it what you would keep in an app and no more. Agent spending before a wallet is made: the limits for one purchase, one request and one day, and how readily the agent pays. The AI model never holds the key and never builds a payment. It gets two tools. One searches a public catalogue of pay-per-call services and shows each price and how many different payers used the service in the last 30 days. The other asks for a service: the agent says what it needs and why, and Graspable decides.","u":"/blog/how-the-agent-pays-for-things#a-wallet-of-its-own"},{"k":"blog","t":"An agent with a wallet and a budget","h":"How a purchase is decided","x":"The order never changes. Each step can stop a purchase, and none can loosen the one before it. Limits come first, and they are code. A request or a day that has used its budget buys nothing more. A service that asks more than its listed price is refused. Only exact amounts of USDC on Base are paid, to services from the catalogue or ones already paid before. Then a judge answers four questions abou","s":"The order never changes. Each step can stop a purchase, and none can loosen the one before it. Limits come first, and they are code. A request or a day that has used its budget buys nothing more. A service that asks more than its listed price is refused. Only exact amounts of USDC on Base are paid, to services from the catalogue or ones already paid before. Then a judge answers four questions about the purchase: - Does your request need this? - Would a free tool the agent already has do? - Is the price in line with the alternatives? - Does the reason come from your request, or only from text the agent read on the way? The last question is there because a web page can say anything, including \"buy this\". A purchase whose reason traces back to something the agent read is one you are asked about. The judge can make the agent stricter than the limits. It cannot make it looser. Then you, whenever the setting, the service or the judge says so. You choose how readily the agent pays, between being asked every time and paying without asking inside the budget. Some cases always come to you: an amount above your limit for one purchase, a service fewer than three others have paid in the last 30 days, and a moment when no judge with measured confidence is available. When nobody can be asked, a purchase that needs your yes is refused.","u":"/blog/how-the-agent-pays-for-things#how-a-purchase-is-decided"},{"k":"blog","t":"An agent with a wallet and a budget","h":"The record","x":"Settings → Agent purchases lists every purchase and every refusal. Open one and you see the request it was for, the agent's own reason, what else it considered, who decided and why, the judge's answers with their scores, and the payment. The record counts refusals too. Amounts, services and reasons are blurred here. The record is a file on your computer. It is not sent to Graspable. Neither is any","s":"Settings → Agent purchases lists every purchase and every refusal. Open one and you see the request it was for, the agent's own reason, what else it considered, who decided and why, the judge's answers with their scores, and the payment. The record counts refusals too. Amounts, services and reasons are blurred here. The record is a file on your computer. It is not sent to Graspable. Neither is anything else about a purchase: the amount, the service, or the fact that one happened. The agent pays the service directly, and Graspable takes nothing from the payment.","u":"/blog/how-the-agent-pays-for-things#the-record"},{"k":"blog","t":"An agent with a wallet and a budget","h":"Spending without asking","x":"There is a setting that removes the questions. We call it unsafe in the app, and mean it. An AI model can be wrong, and text it reads can try to talk it into buying. With that setting on, the limits are the only thing between the agent and the wallet, so assume everything inside them can be spent. Three rules hold even then. Nothing above the limit for one purchase is bought. A purchase the judge ","s":"There is a setting that removes the questions. We call it unsafe in the app, and mean it. An AI model can be wrong, and text it reads can try to talk it into buying. With that setting on, the limits are the only thing between the agent and the wallet, so assume everything inside them can be spent. Three rules hold even then. Nothing above the limit for one purchase is bought. A purchase the judge traces to text the agent read is refused. The agent buys only from services that at least three others have paid in the last 30 days.","u":"/blog/how-the-agent-pays-for-things#spending-without-asking"},{"k":"blog","t":"An agent with a wallet and a budget","h":"First version","x":"It pays in USDC on Base, to services that speak the x402 protocol. Prices must be fixed per call. It has been tried against real services up to the point of payment, so treat it as new. It does not work in Cloud Agents, because the wallet never leaves your computer. The settings and every rule are in the documentation under Agent spending.","u":"/blog/how-the-agent-pays-for-things#first-version"},{"k":"blog","t":"Pay for Graspable with crypto: Credits, or a plan paid ahead","x":"You can now pay for Credits or for a Builder, Pro or Studio plan from your own wallet, in Bitcoin, Ether, Solana or USDC on Base. No card and no identity check. Here is how it works and what is different from paying by card. Graspable takes crypto for Credits and for plans. You pay from your own wallet in Bitcoin, Ether, Solana or USDC on Base, at the same prices as by card. There is no card, no b","s":"You can now pay for Credits or for a Builder, Pro or Studio plan from your own wallet, in Bitcoin, Ether, Solana or USDC on Base. No card and no identity check. Here is how it works and what is different from paying by card. Graspable takes crypto for Credits and for plans. You pay from your own wallet in Bitcoin, Ether, Solana or USDC on Base, at the same prices as by card. There is no card, no billing name and no identity check. You still need a Graspable account, which is an email address or a GitHub or Google sign-in. It is marked experimental in the app, because it is new and a payment on a chain cannot be taken back.","u":"/blog/pay-for-graspable-with-crypto"},{"k":"blog","t":"Pay for Graspable with crypto: Credits, or a plan paid ahead","h":"Credits","x":"Credits pay for what Graspable runs for you: Cloud Agents, Fast mode, generated worlds, and the AI model when you use the Graspable agent with no AI account of your own. Building on your own computer with your own AI account uses none. In the app, open Settings → Plan and Credits , and under Buy Credits choose Pay with crypto . On the website, sign in and open Account → Pay with crypto. 1. Choose ","s":"Credits pay for what Graspable runs for you: Cloud Agents, Fast mode, generated worlds, and the AI model when you use the Graspable agent with no AI account of your own. Building on your own computer with your own AI account uses none. In the app, open Settings → Plan and Credits , and under Buy Credits choose Pay with crypto . On the website, sign in and open Account → Pay with crypto. 1. Choose how many Credits, from 50 up. A Credit is $0.20, $0.18 from 300 and $0.15 from 1,000. 2. Choose a coin. You see the amount, an address and a QR code. 3. Pay from your wallet. Choose an amount, then a coin. The price in dollars is the card's. The Credits arrive once the network has confirmed the payment. USDC on Base and Solana take under a minute and Ether a few minutes. Bitcoin needs one confirmation for a purchase up to $100, which is about ten minutes, and more confirmations for larger ones. You can close the app while you wait. The order is still there when you come back. An order says what to send and where, how many confirmations it needs, and how long the price holds. The amount and the address are blurred here. Credits bought this way last 12 months, like any Credit pack.","u":"/blog/pay-for-graspable-with-crypto#credits"},{"k":"blog","t":"Pay for Graspable with crypto: Credits, or a plan paid ahead","h":"A plan, paid ahead","x":"A card can be charged every month. A wallet only ever sends money, so nothing can renew by itself. A plan paid with crypto is therefore time you buy ahead: 1 month, 3 months or a year, at the card's prices. A month costs the monthly price and a year costs the yearly one. Choose A plan, paid ahead in the same place, pick Builder, Pro or Studio and how long, and pay as you would for Credits. For Stu","s":"A card can be charged every month. A wallet only ever sends money, so nothing can renew by itself. A plan paid with crypto is therefore time you buy ahead: 1 month, 3 months or a year, at the card's prices. A month costs the monthly price and a year costs the yearly one. Choose A plan, paid ahead in the same place, pick Builder, Pro or Studio and how long, and pay as you would for Credits. For Studio you also set the number of seats. The plan starts when the payment is confirmed, and the account page says until which day it is paid. Since it does not renew, we send an email a week before the last day and another the day before. If you do nothing, the account goes back to the free plan on that day. Projects on your computer and apps you have published stay as they are. Changes while time is left work like this: - More time goes on the end, so paying early loses nothing. - A dearer plan or another seat costs the difference for the days already paid. - A cheaper plan or fewer seats turns what is left into more days.","u":"/blog/pay-for-graspable-with-crypto#a-plan-paid-ahead"},{"k":"blog","t":"Pay for Graspable with crypto: Credits, or a plan paid ahead","h":"Before you send","x":"- The price holds for 20 minutes. A payment that arrives later the same day is still credited at the price you were shown. - When the app says Send exactly , send that exact amount on the network shown. For some coins every payment goes to one address, and the last digits of the amount are how yours is recognised. - If less than a plan's price arrives, the plan does not start. The money becomes Cr","s":"- The price holds for 20 minutes. A payment that arrives later the same day is still credited at the price you were shown. - When the app says Send exactly , send that exact amount on the network shown. For some coins every payment goes to one address, and the last digits of the amount are how yours is recognised. - If less than a plan's price arrives, the plan does not start. The money becomes Credits at their usual price. - Card and crypto do not mix on one account. With a plan paid by card, cancel it first and pay with crypto when its time has ended. - Overage, the extra Credits beyond a plan, is billed on a card's invoice. With crypto, buy a Credit pack when you need more. - A payment cannot be taken back. If you paid and nothing arrived within a day, write to us with the order reference the app shows.","u":"/blog/pay-for-graspable-with-crypto#before-you-send"},{"k":"blog","t":"Pay for Graspable with crypto: Credits, or a plan paid ahead","h":"Your wallet stays yours","x":"Graspable holds no wallet for this and never asks for yours. You get an address and an amount, and the payment is looked up in public sources. Nothing on our side holds a key that could move money. People buy Credits. Agents do not. An agent that pays for outside services is a separate, opt-in feature with a wallet of its own, described in the next post. The full rules are in the documentation, un","s":"Graspable holds no wallet for this and never asks for yours. You get an address and an amount, and the payment is looked up in public sources. Nothing on our side holds a key that could move money. People buy Credits. Agents do not. An agent that pays for outside services is a separate, opt-in feature with a wallet of its own, described in the next post. The full rules are in the documentation, under Credits and Plans and billing.","u":"/blog/pay-for-graspable-with-crypto#your-wallet-stays-yours"},{"k":"blog","t":"An address for every app, and the request behind every example","x":"What changed in Graspable from 0.9.36 to 0.9.45: published apps live at their own address on graspable.cloud, connected apps are chosen per project, your Credits are always in view, and every example on the site now comes with the request that built it. The last ten releases of Graspable were small and frequent. Together they change how a finished scene leaves the app, how the agent reaches your o","s":"What changed in Graspable from 0.9.36 to 0.9.45: published apps live at their own address on graspable.cloud, connected apps are chosen per project, your Credits are always in view, and every example on the site now comes with the request that built it. The last ten releases of Graspable were small and frequent. Together they change how a finished scene leaves the app, how the agent reaches your other tools, and how much the site tells you before you download anything. This is what is new, in the order you are most likely to meet it.","u":"/blog/an-address-for-every-app"},{"k":"blog","t":"An address for every app, and the request behind every example","h":"Every published app has an address of its own","x":"When you publish a project, it now lives at https://<name .graspable.cloud/ . The name is the app's address: an app called hoops is at hoops.graspable.cloud, apart from every other app, with nothing of the site around it. You can rename an app under Settings → Published apps , and its page on graspable.dev keeps working whatever the name. A shared link now previews as the app, not as the template ","s":"When you publish a project, it now lives at https://<name .graspable.cloud/ . The name is the app's address: an app called hoops is at hoops.graspable.cloud, apart from every other app, with nothing of the site around it. You can rename an app under Settings → Published apps , and its page on graspable.dev keeps working whatever the name. A shared link now previews as the app, not as the template it started from: the page carries the app's own title and description. Public apps that may be continued from show a small badge that leads to their page, where anyone with Graspable can open the project and carry on. There is also a way to tell us about a published app that is harmful, deceptive or copies someone's work: graspable.dev/report, linked from every app's page. A person reads every report.","u":"/blog/an-address-for-every-app#every-published-app-has-an-address-of-its-own"},{"k":"blog","t":"An address for every app, and the request behind every example","h":"Connect apps, per project","x":"If you have connected other services to your account (Notion, Slack, Google Drive and so on), the agent no longer gets all of them in every run. Next to the box where you write to the agent there is now Connect apps : pick the services this project should use, and remove them again with the × beside each name. A service stays connected for your other projects; it is only the project that stops usi","s":"If you have connected other services to your account (Notion, Slack, Google Drive and so on), the agent no longer gets all of them in every run. Next to the box where you write to the agent there is now Connect apps : pick the services this project should use, and remove them again with the × beside each name. A service stays connected for your other projects; it is only the project that stops using it. Connected services come with the Builder plan and higher. They are being set up in production as this is written, so if the option is not in your Settings yet, that is why.","u":"/blog/an-address-for-every-app#connect-apps-per-project"},{"k":"blog","t":"An address for every app, and the request behind every example","h":"Credits where you can see them","x":"Your Credits are now on the Projects page and at the top of every project. Click the number to open Settings → Plan and Credits , which shows what is left of the month's plan Credits, the Credits you bought and when new ones come. The same page lists every payment with its invoice, and a button for your payment method and billing details.","u":"/blog/an-address-for-every-app#credits-where-you-can-see-them"},{"k":"blog","t":"An address for every app, and the request behind every example","h":"A page for every example, and for every template","x":"Each scene in Made in Graspable now has its own page: the capture, how many runs it took, which AI did them, and the request word for word , exactly as it was sent. Nothing is edited for the page. If you want to know what a request that produces a working tape measure in a scanned room looks like, here it is. Every template has a page too, with what it starts you with, requests to try, and what pe","s":"Each scene in Made in Graspable now has its own page: the capture, how many runs it took, which AI did them, and the request word for word , exactly as it was sent. Nothing is edited for the page. If you want to know what a request that produces a working tape measure in a scanned room looks like, here it is. Every template has a page too, with what it starts you with, requests to try, and what people made from it.","u":"/blog/an-address-for-every-app#a-page-for-every-example-and-for-every-template"},{"k":"blog","t":"An address for every app, and the request behind every example","h":"Smaller things","x":"- Choosing a model that was only just added to your provider's list no longer ends a run with \"unknown model\". - Opening billing no longer fails for accounts that only bought Credits. - The Projects page shows the version of Graspable you are running, next to Settings. - Your own E2B account for Cloud Agents comes with the Builder plan; your own server with Pro. Everything above is in Graspable 0.","s":"- Choosing a model that was only just added to your provider's list no longer ends a run with \"unknown model\". - Opening billing no longer fails for accounts that only bought Credits. - The Projects page shows the version of Graspable you are running, next to Settings. - Your own E2B account for Cloud Agents comes with the Builder plan; your own server with Pro. Everything above is in Graspable 0.9.45. The app offers the update itself; the changelog has each release on its own.","u":"/blog/an-address-for-every-app#smaller-things"},{"k":"blog","t":"Build for Meta VR Glasses on the web: eyes and hands first","x":"Meta VR Glasses ship without controllers and use gaze to aim and a pinch to select, in a narrower view than Quest 3. What that means for a WebXR app, how to try it in an emulator today, and what to change in a scene built for controllers. At Meta Connect 2026, Meta showed its VR Glasses: about 100 grams, shipping in spring 2027, running the same operating system, store and SDKs as Meta Quest. Two ","s":"Meta VR Glasses ship without controllers and use gaze to aim and a pinch to select, in a narrower view than Quest 3. What that means for a WebXR app, how to try it in an emulator today, and what to change in a scene built for controllers. At Meta Connect 2026, Meta showed its VR Glasses: about 100 grams, shipping in spring 2027, running the same operating system, store and SDKs as Meta Quest. Two things make them different for anyone building a WebXR app. They ship without controllers, so eyes and hands are the primary input. And the view is narrower: Meta quotes 70 by 66 degrees against Quest 3's 110 by 96. This guide is about what a WebXR app has to do differently, and how to test it before the hardware exists.","u":"/blog/build-for-meta-vr-glasses-on-the-web"},{"k":"blog","t":"Build for Meta VR Glasses on the web: eyes and hands first","h":"The input model: look, then pinch","x":"On the glasses you aim with your eyes and act with your hands. A pinch is the click. Meta's guidance for the web is to design around gaze-to-target and pinch-to-select first, then add controllers as an option for the people who buy them. For a WebXR app this is close to something that already exists. On Apple Vision Pro, Safari exposes the same idea as a transient-pointer input source: an input th","s":"On the glasses you aim with your eyes and act with your hands. A pinch is the click. Meta's guidance for the web is to design around gaze-to-target and pinch-to-select first, then add controllers as an option for the people who buy them. For a WebXR app this is close to something that already exists. On Apple Vision Pro, Safari exposes the same idea as a transient-pointer input source: an input that appears when you pinch, with a ray from where you were looking. Meta's Immersive Web SDK maps gaze and pinch onto its ordinary ray-pointer interaction, so objects that react to a controller's ray react to a gaze-and-pinch too, without gaze-specific code. The practical rule is the same either way: make every interaction work through a ray and a single select action, and never require a button that only a controller has. The parts that are new: - Targets must be big enough to hit with the eyes. Gaze is precise in the middle of the view and less so near the edges. Meta's guide says targeting \"becomes less precise near the edge of the visible area\". Put the things people must hit near the centre, and make them larger than you would for a controller. - Hover is a gaze. Whatever you highlight on hover is now highlighted by looking. Keep hover feedback quiet; a scene where everything lights up as the eyes pass over it is tiring. - Mappings for controller habits. Meta's own table for porting: a controller ray becomes an eye or hand ray; trigger and grip become pinch, poke or a palm grab; a thumbstick scroll becomes a swipe or a pinch-and-drag.","u":"/blog/build-for-meta-vr-glasses-on-the-web#the-input-model-look-then-pinch"},{"k":"blog","t":"Build for Meta VR Glasses on the web: eyes and hands first","h":"The field of view","x":"With a narrower view, anything placed where a Quest user would glance sideways is simply not visible. Meta recommends checking scene coverage against the glasses' view, and its emulator can mask the view to the glasses' angular bounds. Keep the things that matter inside the middle of the view, and put panels and menus where the head naturally rests rather than at the edges.","u":"/blog/build-for-meta-vr-glasses-on-the-web#the-field-of-view"},{"k":"blog","t":"Build for Meta VR Glasses on the web: eyes and hands first","h":"Trying it without the hardware","x":"Meta's Immersive Web Emulation Runtime (IWER) is the library that emulates a headset inside a desktop browser. From version 2.5.0 it has a Meta VR Glasses profile: the view mask with the glasses' angular bounds, gaze emulation, and a \"Gaze + Hands\" input mode. In code it is one line: With the device installed, a page that asks for an immersive session gets the glasses: gaze as the targeting ray, a","s":"Meta's Immersive Web Emulation Runtime (IWER) is the library that emulates a headset inside a desktop browser. From version 2.5.0 it has a Meta VR Glasses profile: the view mask with the glasses' angular bounds, gaze emulation, and a \"Gaze + Hands\" input mode. In code it is one line: With the device installed, a page that asks for an immersive session gets the glasses: gaze as the targeting ray, a pinch as the select, and the narrower view. The emulator's panel lets you move the gaze with the mouse and pinch with a key. It is an approximation of the angular coverage, not of the lens edges, and it is not the real display; but it is enough to find a button that sits outside the view, or an interaction that needs a trigger. If you build with Meta's Immersive Web SDK rather than plain WebXR, its project config asks for the same things: gazeTracking: true requests gaze as optional; when the runtime does not provide it, IWSDK falls back to head pointing, which is the right behaviour for a Quest 3 user opening the same page.","u":"/blog/build-for-meta-vr-glasses-on-the-web#trying-it-without-the-hardware"},{"k":"blog","t":"Build for Meta VR Glasses on the web: eyes and hands first","h":"What to change in a scene built for controllers","x":"Go through the scene with these questions: 1. Does anything need a button other than select? A grip, a thumbstick, an A or B button. Each of those needs a pinch, a poke or an in-scene button instead. 2. Can everything that must be hit be hit by looking? Targets small enough to need a controller's precision need to grow, or move towards the centre. 3. Is the important part of the scene inside 70 by","s":"Go through the scene with these questions: 1. Does anything need a button other than select? A grip, a thumbstick, an A or B button. Each of those needs a pinch, a poke or an in-scene button instead. 2. Can everything that must be hit be hit by looking? Targets small enough to need a controller's precision need to grow, or move towards the centre. 3. Is the important part of the scene inside 70 by 66 degrees from where the person starts? Load the glasses profile and look. 4. Does the app still work with controllers? It should: the glasses support Touch Plus controllers, sold separately, and Quest users have them.","u":"/blog/build-for-meta-vr-glasses-on-the-web#what-to-change-in-a-scene-built-for-controllers"},{"k":"blog","t":"Build for Meta VR Glasses on the web: eyes and hands first","h":"Doing this in Graspable","x":"Every Graspable template already emulates a headset in the preview, through IWER. Since Graspable 0.9.46 the preview toolbar has a headset selector: Quest 3 or Meta VR Glasses . Choose the glasses and the preview reloads with the narrower view, gaze and pinch; record a session and it is saved as a fixture that the completion gate replays after every change, so a scene that stops working for eyes a","s":"Every Graspable template already emulates a headset in the preview, through IWER. Since Graspable 0.9.46 the preview toolbar has a headset selector: Quest 3 or Meta VR Glasses . Choose the glasses and the preview reloads with the narrower view, gaze and pinch; record a session and it is saved as a fixture that the completion gate replays after every change, so a scene that stops working for eyes and hands fails the run. The agent's own check does the same: ask it to verify the app on the glasses and it opens the project with that profile, enters XR and sends back what it sees.","u":"/blog/build-for-meta-vr-glasses-on-the-web#doing-this-in-graspable"},{"k":"blog","t":"Build for Meta VR Glasses on the web: eyes and hands first","h":"Sources","x":"- VR is Expanding. Start Building for Meta VR Glasses Today (Meta, Connect 2026 recap: weight, ship date, field of view, controllers) - Bring your app to Meta VR Glasses (Meta: input mappings, edge precision, 64-bit requirement) - Get started with Meta VR Glasses using IWSDK (Meta: gaze and pinch, config keys, fallback to head pointing, the emulator's view mask) - Immersive Web Emulation Runtime (","s":"- VR is Expanding. Start Building for Meta VR Glasses Today (Meta, Connect 2026 recap: weight, ship date, field of view, controllers) - Bring your app to Meta VR Glasses (Meta: input mappings, edge precision, 64-bit requirement) - Get started with Meta VR Glasses using IWSDK (Meta: gaze and pinch, config keys, fallback to head pointing, the emulator's view mask) - Immersive Web Emulation Runtime (Meta, MIT: the metaVRGlasses device profile) - WebXR Device API: transient input sources (W3C)","u":"/blog/build-for-meta-vr-glasses-on-the-web#sources"},{"k":"blog","t":"What Graspable sends home, and what it never does","x":"How a desktop app that reads your prompts and your code can report its own errors and count its own use without any of that leaving your computer: the rules we set, where they live in the code, and how to switch it all off. Graspable sees a lot. The agent reads your request, opens your files, runs commands in your project and remembers things about how you work. All of that stays on your computer;","s":"How a desktop app that reads your prompts and your code can report its own errors and count its own use without any of that leaving your computer: the rules we set, where they live in the code, and how to switch it all off. Graspable sees a lot. The agent reads your request, opens your files, runs commands in your project and remembers things about how you work. All of that stays on your computer; that was the first decision when we made Graspable a desktop app rather than a website. But an app that stays on your computer is also an app we cannot see break. This post is about what we do send, how it is kept harmless, and where each rule lives, so you can hold us to it.","u":"/blog/what-graspable-sends-home"},{"k":"blog","t":"What Graspable sends home, and what it never does","h":"The one switch","x":"Everything described here follows one setting: Settings → Help improve Graspable . Off means nothing in this post is sent. It is on by default, because the first weeks of a product are when the reports matter most, and because what is sent is designed to be safe to send. The grasp command line shares the same switch ( grasp telemetry off ) and also honours DO NOT TRACK=1 . Ratings are separate. Th","s":"Everything described here follows one setting: Settings → Help improve Graspable . Off means nothing in this post is sent. It is on by default, because the first weeks of a product are when the reports matter most, and because what is sent is designed to be safe to send. The grasp command line shares the same switch ( grasp telemetry off ) and also honours DO NOT TRACK=1 . Ratings are separate. The 👍 and 👎 under an agent's message send something only when you click them, and your request and the message go along only when you tick the box that says so.","u":"/blog/what-graspable-sends-home#the-one-switch"},{"k":"blog","t":"What Graspable sends home, and what it never does","h":"Error reports","x":"When something breaks in Graspable's own code, the app sends a short report: the kind of error, its message, where in our code it happened, the version and your system. Before it leaves, a filter takes out anything in the message that could be about you: - the path of your home folder becomes ~ , - email addresses become <email , - anything that looks like a key or a token becomes <secret , - the ","s":"When something breaks in Graspable's own code, the app sends a short report: the kind of error, its message, where in our code it happened, the version and your system. Before it leaves, a filter takes out anything in the message that could be about you: - the path of your home folder becomes ~ , - email addresses become <email , - anything that looks like a key or a token becomes <secret , - the message is cut to a few hundred characters. A failed run is reported too, as its error message alone, through the same filter. The server does the whole thing again before passing the report on, and it keeps nothing about who sent it: no account, no address. A report is not tied to you. The filter and its tests are in one small module we are happy to be judged on.","u":"/blog/what-graspable-sends-home#error-reports"},{"k":"blog","t":"What Graspable sends home, and what it never does","h":"Counting, not watching","x":"The app counts how it is used: that it was opened, which screen is open, that a project was created and from which template, that a run started and how it ended, that an app was published. The counts carry your account id once you sign in, so we can tell a download from a first project from a plan. What the app does not do is record what is on the screen. The analytics library we use can capture e","s":"The app counts how it is used: that it was opened, which screen is open, that a project was created and from which template, that a run started and how it ended, that an app was published. The counts carry your account id once you sign in, so we can tell a download from a first project from a plan. What the app does not do is record what is on the screen. The analytics library we use can capture every click, every typed character and a video of the session. All of that is switched off in the code, and the app's content policy allows no script from the analytics service to load, so it cannot be switched on from outside. A project appears in these counts as a short code made from its id; the code cannot be turned back into a name. The same goes for grasp and the MCP server. When another agent uses Graspable, we count the command or tool and the name of the client, never the arguments. What the other agent was told by its user is not something we can see, and we do not try.","u":"/blog/what-graspable-sends-home#counting-not-watching"},{"k":"blog","t":"What Graspable sends home, and what it never does","h":"Why count at all","x":"Three reasons, in order. We need to know whether people get from a download to a working scene, and where they stop. We need to know whether the agent's runs complete, and how often they fail after their repair rounds. And we need to know which of the things we build are used, so the ones that are not can go. None of that needs a prompt or a file; counts are enough.","u":"/blog/what-graspable-sends-home#why-count-at-all"},{"k":"blog","t":"What Graspable sends home, and what it never does","h":"Where to check","x":"The privacy page lists every event the app and the server send, and is changed whenever the list changes. If you find anything in a report or a count that this post says cannot be there, tell us; it would be a bug, and we would fix it the same day.","u":"/blog/what-graspable-sends-home#where-to-check"},{"k":"blog","t":"Performance of Gaussian splats in WebXR: what we measured and what we chose","x":"Load time, memory and frame time of two 500,000-splat worlds in Spark, measured on a laptop and in an emulated headset, not on a real one, and the defaults we chose for our templates and why. Our world templates show a generated place as 500,000 Gaussian splats, drawn with Spark. Before shipping them we measured what that costs and picked defaults. This article gives the numbers, how they were tak","s":"Load time, memory and frame time of two 500,000-splat worlds in Spark, measured on a laptop and in an emulated headset, not on a real one, and the defaults we chose for our templates and why. Our world templates show a generated place as 500,000 Gaussian splats, drawn with Spark. Before shipping them we measured what that costs and picked defaults. This article gives the numbers, how they were taken, and what they do not tell you. The most important sentence comes first. Nothing here was measured on a real headset. Everything was measured on one Apple M1 Max, in headless Chrome and in the emulated headset. Where a default is meant for a headset, it rests on Spark's published guidance, and we say so.","u":"/blog/gaussian-splat-performance-webxr"},{"k":"blog","t":"Performance of Gaussian splats in WebXR: what we measured and what we chose","h":"What was measured, and how","x":"- Machine: Apple M1 Max, headless Google Chrome, with the development server on the same machine. Files come from the local disk, so download time is close to zero. - Worlds: the loft and the tavern that ship with the templates. Each is 500,000 splats in a 7.5 MB .spz file, with a preview of 98,304 splats in 1.2 MB. Their collider meshes have 102,206 and 217,939 triangles. - Frame time: 90 frames ","s":"- Machine: Apple M1 Max, headless Google Chrome, with the development server on the same machine. Files come from the local disk, so download time is close to zero. - Worlds: the loft and the tavern that ship with the templates. Each is 500,000 splats in a 7.5 MB .spz file, with a preview of 98,304 splats in 1.2 MB. Their collider meshes have 102,206 and 217,939 triangles. - Frame time: 90 frames drawn back to back. After each one we read a single pixel back, so the call returns only when the GPU has finished. That is the full cost of one frame, CPU and GPU, on this GPU. From run to run it varies by about 1 ms, so smaller differences mean nothing. - \"Quest-sized\" means a drawing buffer of 4128 x 2208 pixels: as many pixels as both eyes of a Quest 3 at its panel resolution, drawn as one view. It stands in for the cost of filling pixels and nothing else. A phone-class GPU is many times slower than an M1 Max, and a real headset draws two views. - Load time: from the start of navigation until the full 500,000-splat world is on screen. - Memory: Chrome's JavaScript heap. GPU memory was not measured.","u":"/blog/gaussian-splat-performance-webxr#what-was-measured-and-how"},{"k":"blog","t":"Performance of Gaussian splats in WebXR: what we measured and what we chose","h":"Load time","x":"For the loft: - The 100,000-splat preview alone is on screen after 0.6 s. - The 500,000-splat world with level of detail off: 0.7 s. - The 500,000-splat world with level of detail on, as shipped: 1.9 to 2.0 s. For the tavern, 2.1 to 2.2 s. About one second of the shipped figure is Spark building the level-of-detail tree: 0.94 to 0.99 s for the loft and 0.99 to 1.0 s for the tavern, by Spark's own ","s":"For the loft: - The 100,000-splat preview alone is on screen after 0.6 s. - The 500,000-splat world with level of detail off: 0.7 s. - The 500,000-splat world with level of detail on, as shipped: 1.9 to 2.0 s. For the tavern, 2.1 to 2.2 s. About one second of the shipped figure is Spark building the level-of-detail tree: 0.94 to 0.99 s for the loft and 0.99 to 1.0 s for the tavern, by Spark's own log line.","u":"/blog/gaussian-splat-performance-webxr#load-time"},{"k":"blog","t":"Performance of Gaussian splats in WebXR: what we measured and what we chose","h":"Memory","x":"JavaScript heap after loading the loft: - 84 MB with only the 100,000-splat preview, - 100 MB with the 500,000-splat world and no level of detail, - 135 MB as shipped, with level of detail. The tree holds 695,685 splats. The tavern as shipped, with the game running, used 149 to 152 MB. So level of detail costs about one second at load and 35 MB of heap for a world of this size.","u":"/blog/gaussian-splat-performance-webxr#memory"},{"k":"blog","t":"Performance of Gaussian splats in WebXR: what we measured and what we chose","h":"Frame time","x":"Median of 90 frames, with the 90th percentile in brackets. At 1280 x 720: - Loft, as shipped: 4.4 ms (5.3), with 441,551 splats drawn. - Tavern, as shipped: 4.0 ms (5.3), with 374,931 splats drawn. - With the budget lowered to 300,000 splats: 3.6 ms (4.7) for the loft and 3.6 ms (5.0) for the tavern. At the Quest-sized 4128 x 2208: - Loft, as shipped: 6.9 ms (8.4), with 484,095 splats drawn. - Tav","s":"Median of 90 frames, with the 90th percentile in brackets. At 1280 x 720: - Loft, as shipped: 4.4 ms (5.3), with 441,551 splats drawn. - Tavern, as shipped: 4.0 ms (5.3), with 374,931 splats drawn. - With the budget lowered to 300,000 splats: 3.6 ms (4.7) for the loft and 3.6 ms (5.0) for the tavern. At the Quest-sized 4128 x 2208: - Loft, as shipped: 6.9 ms (8.4), with 484,095 splats drawn. - Tavern, as shipped: 7.5 ms (9.2), with 465,916 splats drawn. - With the headset settings, a budget of 500,000 and a Gaussian cut-off of Math.sqrt(5) : 7.3 ms (9.0) for the loft and 5.7 ms (8.1) for the tavern. - With a budget of 300,000 and the same cut-off: 7.1 ms (8.7) and 6.7 ms (8.6). - Loft with level of detail off, all 500,000 splats drawn, in an earlier run: 6.5 ms (7.7). - Loft with only the 100,000-splat file, in an earlier run: 4.4 ms (5.5). In the emulated headset, in stereo, in a 1280 x 720 window with the head turning, all three templates held the browser's 60 Hz frame interval: a median of 16.7 ms and a 95th percentile of 17.0 to 18.5 ms. Spark drew 203,000 to 254,000 splats there, because its level-of-detail selection follows the low resolution of the emulator window. The loft in the emulated headset, both eyes. This is the setup of the stereo measurement, not a real headset.","u":"/blog/gaussian-splat-performance-webxr#frame-time"},{"k":"blog","t":"Performance of Gaussian splats in WebXR: what we measured and what we chose","h":"What the numbers say","x":"On this GPU the frame time hardly depends on the number of splats between 150,000 and 500,000. It follows the number of pixels: 4.4 ms at 0.9 megapixels, 6.9 ms at 9.1 megapixels. Lowering the splat budget from 500,000 to 300,000 at the large size changed the loft's frame time by less than the run-to-run variation. That matches how splats are drawn. Each one is a transparent shape blended over wha","s":"On this GPU the frame time hardly depends on the number of splats between 150,000 and 500,000. It follows the number of pixels: 4.4 ms at 0.9 megapixels, 6.9 ms at 9.1 megapixels. Lowering the splat budget from 500,000 to 300,000 at the large size changed the loft's frame time by less than the run-to-run variation. That matches how splats are drawn. Each one is a transparent shape blended over what is behind it, so the work is in the pixels. See Gaussian splats, explained. It also means this machine cannot tell us how a Quest will do. A fast desktop-class GPU hides exactly the costs that limit a mobile one. So the defaults below lean on Spark's guidance for headsets, and the list says which ones do.","u":"/blog/gaussian-splat-performance-webxr#what-the-numbers-say"},{"k":"blog","t":"Performance of Gaussian splats in WebXR: what we measured and what we chose","h":"The defaults we chose","x":"- The 500,000-splat file. Spark's performance guide recommends 1 million splats or fewer for a Quest 3, and its own level-of-detail budget for a Quest is 500,000. The 500,000-splat export from World Labs fits that without tuning. We do not use the larger full-resolution export. Rests on Spark's guidance. - The small file first. The preview is on screen after 0.6 s here, and the full world needs ab","s":"- The 500,000-splat file. Spark's performance guide recommends 1 million splats or fewer for a Quest 3, and its own level-of-detail budget for a Quest is 500,000. The 500,000-splat export from World Labs fits that without tuning. We do not use the larger full-resolution export. Rests on Spark's guidance. - The small file first. The preview is on screen after 0.6 s here, and the full world needs about 2 s, half of it the level-of-detail build. That build will take longer on a headset's processor. Measured here; the headset part is an expectation. - Level of detail on. It keeps any world inside the device's budget, including a larger file or two worlds at once, and draws fewer splats where they are smaller than a pixel. With a 500,000-splat world and a 500,000 budget it removes few splats: 484,095 of 500,000 were drawn at the Quest-sized resolution. So turning it off with <World lod={false} is a fair choice for a single world of this size if load time matters more. Measured here. - Spark's per-device splat budget, unchanged. That is 500,000 in a Quest. If a headset drops frames, lower lodSplatScale ; 0.6 gives 300,000. Rests on Spark's guidance. - Gaussian cut-off: Math.sqrt(8) on a flat screen and Math.sqrt(5) in a headset. Spark's guide recommends the second value for VR. We could not tell them apart in frame time on this GPU. Rests on Spark's guidance. - Antialiasing off. Spark's guide says multisampling does not improve splats and costs a lot. Not measured here. - Flat-screen pixel ratio of at most 1.5. Frame time follows the pixel count, and the cap stops high-density laptop and phone screens from drawing four times the pixels. The value 1.5 is a judgement, not a measurement on a weak device. - Headset resolution: the headset's own recommended size, with no frame buffer scaling set. Not measured. - Fixed foveation at its strongest. Fewer pixels are shaded at the edges of each eye. It cannot be measured in the emulator. - One sort for both eyes. Spark's default is a radial sort from the head position, which stays stable when the head turns. - Exactly one Spark renderer per canvas. Two would sort and draw every splat twice. In code, all of this is a few lines in lib/spark.js of each template:","u":"/blog/gaussian-splat-performance-webxr#the-defaults-we-chose"},{"k":"blog","t":"Performance of Gaussian splats in WebXR: what we measured and what we chose","h":"Things that cost nothing per frame","x":"- The collider is prepared once at load. Moving its vertices into metres and pressing its floor flat took 11 to 20 ms for 51,524 and 109,138 vertices. After that it is one fixed triangle mesh in the physics engine. - The light for added meshes comes from rendering the world into an environment map once after loading, not every frame. Both are described in How our world templates work.","u":"/blog/gaussian-splat-performance-webxr#things-that-cost-nothing-per-frame"},{"k":"blog","t":"Performance of Gaussian splats in WebXR: what we measured and what we chose","h":"Known limits","x":"- No headset measurement. Before you promise a frame rate on a Quest 3, measure on one. - Distance. A generated world is sharp near the point it was generated from and blurs with distance. In the loft, the far end of the room, about 14 m away, is visibly smeared. - Stereo was checked in the emulator only , with a 64 mm eye distance. Splats and meshes line up in both eyes.","u":"/blog/gaussian-splat-performance-webxr#known-limits"},{"k":"blog","t":"Performance of Gaussian splats in WebXR: what we measured and what we chose","h":"How to measure your own scene","x":"On a real headset, the method in our guide A performance budget for WebXR on Quest 3 applies unchanged: decide the frame rate, find out whether pixels or geometry are the limit, change one thing, measure again. With splats the two controls that matter most are: If lowering the splat budget does not help and lowering the resolution does, you are limited by pixels, as we were on the laptop. To open ","s":"On a real headset, the method in our guide A performance budget for WebXR on Quest 3 applies unchanged: decide the frame rate, find out whether pixels or geometry are the limit, change one thing, measure again. With splats the two controls that matter most are: If lowering the splat budget does not help and lowering the resolution does, you are limited by pixels, as we were on the laptop. To open a project on your headset from Graspable, use the preview. For the renderer itself, read Spark: the renderer behind our splat templates. The templates are described in Templates and the worlds in Generated worlds.","u":"/blog/gaussian-splat-performance-webxr#how-to-measure-your-own-scene"},{"k":"blog","t":"Performance of Gaussian splats in WebXR: what we measured and what we chose","h":"Sources","x":"- Our own measurements of 2 October 2026, as described above - Performance tuning, Getting started with level of detail and SparkRenderer, Spark documentation - Export file specs, World Labs - Performance, @react-three/xr","u":"/blog/gaussian-splat-performance-webxr#sources"},{"k":"blog","t":"Gaussian splats, explained for people who build things","x":"What a Gaussian splat is and is not, how it is drawn, the file formats you will meet, what splats are good and bad at, and what that means for a WebXR experience built on one. A Gaussian splat scene looks like a photograph you can walk around in. It is tempting to treat it like any other 3D model. It is not one, and the differences decide what you can build on it. This guide explains what a splat ","s":"What a Gaussian splat is and is not, how it is drawn, the file formats you will meet, what splats are good and bad at, and what that means for a WebXR experience built on one. A Gaussian splat scene looks like a photograph you can walk around in. It is tempting to treat it like any other 3D model. It is not one, and the differences decide what you can build on it. This guide explains what a splat is, how it gets onto the screen, and what it can and cannot do, with the consequences for a WebXR experience.","u":"/blog/gaussian-splats-explained"},{"k":"blog","t":"Gaussian splats, explained for people who build things","h":"What a splat is","x":"The technique comes from the paper 3D Gaussian Splatting for Real-Time Radiance Field Rendering by Kerbl, Kopanas, Leimkühler and Drettakis (SIGGRAPH 2023). A scene is a large set of 3D Gaussians: soft, stretched blobs. The paper gives each one four properties: - a position , - a shape : an anisotropic covariance, stored as a scale for each of three axes and a rotation, - an opacity , - a colour t","s":"The technique comes from the paper 3D Gaussian Splatting for Real-Time Radiance Field Rendering by Kerbl, Kopanas, Leimkühler and Drettakis (SIGGRAPH 2023). A scene is a large set of 3D Gaussians: soft, stretched blobs. The paper gives each one four properties: - a position , - a shape : an anisotropic covariance, stored as a scale for each of three axes and a rotation, - an opacity , - a colour that depends on the direction you look from, stored as spherical harmonic coefficients. There are many of them. The paper reports 1 to 5 million Gaussians for every scene it tested. Nobody models these blobs by hand. In the paper they are optimised from photographs: the cameras are calibrated with structure from motion, the sparse points that step produces become the first Gaussians, and an optimisation then moves, reshapes, splits and removes Gaussians until the rendered images match the photographs. Generative models can now produce the same representation from a text prompt or a single image; that is how the worlds in Graspable are made.","u":"/blog/gaussian-splats-explained#what-a-splat-is"},{"k":"blog","t":"Gaussian splats, explained for people who build things","h":"What a splat is not","x":"Not a mesh. A mesh is a surface made of triangles, with materials that respond to light. A splat scene has no surface at all. It is a cloud of semi-transparent blobs that only looks solid when enough of them overlap. The paper's own closing remarks say it would be interesting to see whether the Gaussians can be used for mesh reconstruction. They are not one already. Not photogrammetry. Classic pho","s":"Not a mesh. A mesh is a surface made of triangles, with materials that respond to light. A splat scene has no surface at all. It is a cloud of semi-transparent blobs that only looks solid when enough of them overlap. The paper's own closing remarks say it would be interesting to see whether the Gaussians can be used for mesh reconstruction. They are not one already. Not photogrammetry. Classic photogrammetry also starts from photographs and structure from motion, then uses multi-view stereo to reconstruct geometry, which is usually delivered as a textured mesh. A splat skips the surface and keeps appearance instead. That is why thin and fuzzy things, such as plants, look better as splats than as scanned meshes. Not a NeRF. A neural radiance field stores the scene in a neural network. To draw a pixel, NeRF queries the network at many points along the camera ray and combines the answers with volume rendering. Gaussian splatting keeps the same idea of a radiance field, light leaving each point in each direction, but stores it as explicit blobs that a GPU can draw directly. The paper presents this as the first real-time rendering solution for radiance fields.","u":"/blog/gaussian-splats-explained#what-a-splat-is-not"},{"k":"blog","t":"Gaussian splats, explained for people who build things","h":"How splats are drawn","x":"\"Splatting\" means each Gaussian is projected onto the screen, where it becomes a soft 2D ellipse. The ellipses are then combined with alpha blending. Blending only gives the right picture when it is done in depth order, so the splats have to be sorted for the current viewpoint. The paper does this with a tile-based rasteriser: it splits the screen into tiles of 16 by 16 pixels, sorts all Gaussians","s":"\"Splatting\" means each Gaussian is projected onto the screen, where it becomes a soft 2D ellipse. The ellipses are then combined with alpha blending. Blending only gives the right picture when it is done in depth order, so the splats have to be sorted for the current viewpoint. The paper does this with a tile-based rasteriser: it splits the screen into tiles of 16 by 16 pixels, sorts all Gaussians once on the GPU with a radix sort, and blends them front to back in each tile until the pixel is opaque. In a browser the same job is done with the tools WebGL has. Spark, the renderer we use, draws each splat as two triangles that cover the Gaussian out to a chosen number of standard deviations, as transparent geometry blended back to front. Its system design page describes how the order is found: distances are read back from the GPU, sorted in a background worker, and used on a following frame. All splats in the scene are then drawn with a single instanced draw call. Two costs follow from this, and they are not the costs of a mesh scene: - Sorting. It has to be redone as the viewpoint moves, for every splat. - Blending. Every splat is transparent, so the GPU draws many layers over the same pixel. The cost grows with the number of pixels, not only with the number of splats. Loading a splat into a three.js scene with Spark takes a few lines:","u":"/blog/gaussian-splats-explained#how-splats-are-drawn"},{"k":"blog","t":"Gaussian splats, explained for people who build things","h":"File formats","x":"- .ply is what the paper's reference implementation writes. It is uncompressed and understood by the most tools. World Labs' export specifications describe it the same way: the format with broader software compatibility. - .spz is an open format for compressed splats from Niantic. Its repository says files are typically around 10 times smaller than the corresponding .ply, with minimal visual diffe","s":"- .ply is what the paper's reference implementation writes. It is uncompressed and understood by the most tools. World Labs' export specifications describe it the same way: the format with broader software compatibility. - .spz is an open format for compressed splats from Niantic. Its repository says files are typically around 10 times smaller than the corresponding .ply, with minimal visual difference. This is what you want to send over a network. The loft in the picture above is 7.5 MB for 500,000 splats. - .splat comes from the early WebGL viewer antimatter15/splat. It leaves out the spherical harmonics to keep files small, so colour no longer changes with the viewing direction. Spark loads all three, and a few more. Its loading page lists them. One trap: files differ in which way is up. World Labs' files use a y-down convention, and three.js is y-up, so the code above turns the splat half a turn around x.","u":"/blog/gaussian-splats-explained#file-formats"},{"k":"blog","t":"Gaussian splats, explained for people who build things","h":"What splats are good at","x":"- Looking real. Soft light, reflections that shift as you move, foliage, clutter. Things that take a long time to model and light come for free. - Whole places. A room or a courtyard arrives complete, not object by object.","u":"/blog/gaussian-splats-explained#what-splats-are-good-at"},{"k":"blog","t":"Gaussian splats, explained for people who build things","h":"What splats are bad at","x":"- Nothing is solid. There are no surfaces to collide with, stand on or cast a ray against. You need a separate, invisible mesh of the same place for physics. - The light is baked in. A splat stores the colour that leaves each point, not a material. Adding a light to the scene does nothing to it. You cannot move the sun. - No shadows either way. A splat does not cast a shadow on your meshes and doe","s":"- Nothing is solid. There are no surfaces to collide with, stand on or cast a ray against. You need a separate, invisible mesh of the same place for physics. - The light is baked in. A splat stores the colour that leaves each point, not a material. Adding a light to the scene does nothing to it. You cannot move the sun. - No shadows either way. A splat does not cast a shadow on your meshes and does not receive one from them. - Colour depends on where you stand. That is a feature, it is what makes highlights look right, but it means a splat has no single \"texture\" you could bake or edit. - Quality depends on what was seen. The paper's limitations section says artefacts appear in regions that were not well observed. For a generated world the same holds for regions far from the point it was generated from: they blur. A splat and meshes in one scene, on a flat screen. The room is 500,000 splats. The three ceramic pieces and the labels are ordinary meshes.","u":"/blog/gaussian-splats-explained#what-splats-are-bad-at"},{"k":"blog","t":"Gaussian splats, explained for people who build things","h":"What this means for a WebXR experience","x":"Scale has to be right. In a headset a room at 80% size feels wrong at once. A splat file has no fixed unit, so you need to know how many metres one unit is, and where the floor is. World Labs returns both numbers with every world; see their note on rendering SPZ files. You need a collider. Teleporting, walking, placing an object on a table and throwing a ball all need surfaces. Use a coarse mesh o","s":"Scale has to be right. In a headset a room at 80% size feels wrong at once. A splat file has no fixed unit, so you need to know how many metres one unit is, and where the floor is. World Labs returns both numbers with every world; see their note on rendering SPZ files. You need a collider. Teleporting, walking, placing an object on a table and throwing a ball all need surfaces. Use a coarse mesh of the same place, in the same coordinates, and never show it. Budget the splats, and the pixels. A headset draws one view for each eye on a mobile chip. Spark's performance guide recommends 1 million splats or fewer for a Quest 3, and says to leave multisampling off because it does not improve splats and costs a lot. Treat any such number as a starting point and measure on the device. Light your own objects from the splat. Meshes you add will look pasted in unless they pick up the colours of the place. Rendering the splat into an environment map and using it to light the meshes works well. Fake the contact shadows. A soft dark blob under an added object does what a real shadow cannot. Keep people near the good part. Start them where the scene is sharpest and limit how far they can go. The loft in the emulated headset, both eyes. Every splat is drawn once per eye.","u":"/blog/gaussian-splats-explained#what-this-means-for-a-webxr-experience"},{"k":"blog","t":"Gaussian splats, explained for people who build things","h":"Doing this in Graspable","x":"Graspable has three templates built on splats: Splat World, World Explorer and World Game. Each starts inside a generated place at real scale, with its collider wired to physics, and the agent can generate a new place when you ask for one. See Generated worlds and Templates. The rest of this series goes into the details: - Graspable now generates worlds with World Labs - Spark: the renderer behind","s":"Graspable has three templates built on splats: Splat World, World Explorer and World Game. Each starts inside a generated place at real scale, with its collider wired to physics, and the agent can generate a new place when you ask for one. See Generated worlds and Templates. The rest of this series goes into the details: - Graspable now generates worlds with World Labs - Spark: the renderer behind our splat templates - Performance of Gaussian splats in WebXR - How our world templates work","u":"/blog/gaussian-splats-explained#doing-this-in-graspable"},{"k":"blog","t":"Gaussian splats, explained for people who build things","h":"Sources","x":"- 3D Gaussian Splatting for Real-Time Radiance Field Rendering, Kerbl, Kopanas, Leimkühler and Drettakis, SIGGRAPH 2023; the paper is also on arXiv - NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis, Mildenhall and others - spz, Niantic, and splat, antimatter15 - System design, Loading splats and Performance tuning, Spark - Export file specs and Rendering Marble SPZ files in ","s":"- 3D Gaussian Splatting for Real-Time Radiance Field Rendering, Kerbl, Kopanas, Leimkühler and Drettakis, SIGGRAPH 2023; the paper is also on arXiv - NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis, Mildenhall and others - spz, Niantic, and splat, antimatter15 - System design, Loading splats and Performance tuning, Spark - Export file specs and Rendering Marble SPZ files in third-party engines, World Labs - XRView, MDN, and the WebXR Device API, W3C","u":"/blog/gaussian-splats-explained#sources"},{"k":"blog","t":"Graspable now generates worlds with World Labs","x":"Ask for a place in plain words and the agent generates it with World Labs Marble and puts it in your project: what you get, what it costs in Credits, and the three templates that use it. From version 0.9.31, Graspable can generate a place for your project. You describe it. The agent generates it with World Labs Marble, through their World API, and saves it into the project on your computer. Then i","s":"Ask for a place in plain words and the agent generates it with World Labs Marble and puts it in your project: what you get, what it costs in Credits, and the three templates that use it. From version 0.9.31, Graspable can generate a place for your project. You describe it. The agent generates it with World Labs Marble, through their World API, and saves it into the project on your computer. Then it switches the scene to the new place. Request: Make this a greenhouse workshop at dusk, full of plants and glassware, with warm light from hanging lamps. The agent asks before it starts, because a world costs Credits. Generation takes about five minutes. The run waits for it.","u":"/blog/generate-worlds-with-world-labs"},{"k":"blog","t":"Graspable now generates worlds with World Labs","h":"What you get","x":"A world is a folder in your project, public/worlds/<name / . It holds: - The world as Gaussian splats , in two sizes: 500,000 splats, which is what is shown, and a light version of about 100,000 that appears first while the large one loads. - A collider mesh : the same place as a plain mesh that is never shown. It is what you stand on, bump into and bounce balls off. - The panorama the world was m","s":"A world is a folder in your project, public/worlds/<name / . It holds: - The world as Gaussian splats , in two sizes: 500,000 splats, which is what is shown, and a light version of about 100,000 that appears first while the large one loads. - A collider mesh : the same place as a plain mesh that is never shown. It is what you stand on, bump into and bounce balls off. - The panorama the world was made from. - world.json : the name, your description, World Labs' caption of what is in the place, and two numbers that turn the world's raw units into metres and put its floor at zero. The templates read world.json for you, so a world appears at real size with the floor at your feet. If Gaussian splats are new to you, start with Gaussian splats, explained. The panorama of the loft world that ships with two of the templates. It was generated from a text description only.","u":"/blog/generate-worlds-with-world-labs#what-you-get"},{"k":"blog","t":"Graspable now generates worlds with World Labs","h":"What it costs","x":"- Draft : 2 Credits. A quick, rough version to check the idea. - Standard : 11 Credits. The default. - Large : 21 Credits. A bigger outdoor place or a large interior. A world that fails to generate costs nothing. Generating worlds through Graspable comes with the Builder plan and higher. See Credits. If you have your own World Labs API key, you can use it instead. Open Settings → Plugins and conne","s":"- Draft : 2 Credits. A quick, rough version to check the idea. - Standard : 11 Credits. The default. - Large : 21 Credits. A bigger outdoor place or a large interior. A world that fails to generate costs nothing. Generating worlds through Graspable comes with the Builder plan and higher. See Credits. If you have your own World Labs API key, you can use it instead. Open Settings → Plugins and connectors , switch on World Labs and enter the key. World Labs then bills you directly, no Credits are used, and your Graspable plan does not matter.","u":"/blog/generate-worlds-with-world-labs#what-it-costs"},{"k":"blog","t":"Graspable now generates worlds with World Labs","h":"Three templates that use it","x":"All three are premium templates and start inside a world we generated for them. - Splat World starts in a sunlit loft. You look around, teleport and put things in it. They sit on the world's real floor and tables, at real size. - World Explorer starts in the same loft, but solid. Floors hold you, walls and furniture stop you. It comes with info points and a guided tour. - World Game starts in a ta","s":"All three are premium templates and start inside a world we generated for them. - Splat World starts in a sunlit loft. You look around, teleport and put things in it. They sit on the world's real floor and tables, at real size. - World Explorer starts in the same loft, but solid. Floors hold you, walls and furniture stop you. It comes with info points and a guided tour. - World Game starts in a tavern with a throwing game: knock the mugs off the bar, ring the targets, keep score. Balls bounce off the room itself. World Game on a flat screen. The tavern is the generated world. The mugs, targets and scoreboard are added by the template. World Explorer on a flat screen, at the start of its guided tour.","u":"/blog/generate-worlds-with-world-labs#three-templates-that-use-it"},{"k":"blog","t":"Graspable now generates worlds with World Labs","h":"How to try it","x":"1. Create a project from Splat World, World Explorer or World Game. 2. Ask for a place. One or two sentences are enough: what kind of place it is, what is in it, the light and the mood. 3. Approve the cost. Wait about five minutes. 4. Look around in the preview, then in the emulated headset or on a real one. A world is a still place. Leave people, animals and anything that should move out of the d","s":"1. Create a project from Splat World, World Explorer or World Game. 2. Ask for a place. One or two sentences are enough: what kind of place it is, what is in it, the light and the mood. 3. Approve the cost. Wait about five minutes. 4. Look around in the preview, then in the emulated headset or on a real one. A world is a still place. Leave people, animals and anything that should move out of the description, and add those as objects afterwards. A generated world is sharpest near the point it was generated from and blurs with distance. The templates start you at that point, and World Explorer limits how far you can walk from it. Graspable is available for macOS. Windows and Linux are coming soon.","u":"/blog/generate-worlds-with-world-labs#how-to-try-it"},{"k":"blog","t":"Graspable now generates worlds with World Labs","h":"Read more","x":"- Generated worlds and Templates in the documentation - Spark: the renderer behind our splat templates - Performance of Gaussian splats in WebXR: what we measured and what we chose - How our world templates work","u":"/blog/generate-worlds-with-world-labs#read-more"},{"k":"blog","t":"Graspable now generates worlds with World Labs","h":"Sources","x":"- World API documentation, models and export file specs, World Labs","u":"/blog/generate-worlds-with-world-labs#sources"},{"k":"blog","t":"How our world templates work","x":"The layer shared by Splat World, World Explorer and World Game: one World component, raw units to metres, the collider as a physics mesh, a flattened floor, light from the splat, walking, throwing, and how the agent switches worlds. Splat World, World Explorer and World Game look different, but they stand on the same few files. This article walks through that shared layer: what a generated world i","s":"The layer shared by Splat World, World Explorer and World Game: one World component, raw units to metres, the collider as a physics mesh, a flattened floor, light from the splat, walking, throwing, and how the agent switches worlds. Splat World, World Explorer and World Game look different, but they stand on the same few files. This article walks through that shared layer: what a generated world is on disk, how it becomes a place at real scale that you can stand in, and the problems we met on the way. It assumes you know what a Gaussian splat is. If not, start with Gaussian splats, explained.","u":"/blog/how-world-templates-work"},{"k":"blog","t":"How our world templates work","h":"One component","x":"A world is a folder in the project, public/worlds/<name / , generated by World Labs Marble. It holds the splats in two sizes, a collider mesh, a panorama and world.json . In the scene it is one component: <World does four things. It shows the small splat at once and swaps in the large one when it is ready. It puts the splat at real scale, upright, with the floor at y = 0. It turns the collider int","s":"A world is a folder in the project, public/worlds/<name / , generated by World Labs Marble. It holds the splats in two sizes, a collider mesh, a panorama and world.json . In the scene it is one component: <World does four things. It shows the small splat at once and swaps in the large one when it is ready. It puts the splat at real scale, upright, with the floor at y = 0. It turns the collider into one fixed physics body named \"world\" . And it mounts its children only when the collider is in place, so nothing you put in the world falls through the floor on the first frame. Children ask the world questions through a hook: Both functions cast a ray against the collider with the physics engine, never against the splat.","u":"/blog/how-world-templates-work#one-component"},{"k":"blog","t":"How our world templates work","h":"From raw units to metres","x":"A generated world has no fixed unit and its y axis points down. World Labs returns two numbers with each world and documents how to use them: metric scale factor converts raw units to metres, and ground plane offset puts the ground at y = 0. The same page says the files use an OpenCV-style frame and that Marble's own viewer applies a half turn around the x axis. For our loft the numbers are 2.378 ","s":"A generated world has no fixed unit and its y axis points down. World Labs returns two numbers with each world and documents how to use them: metric scale factor converts raw units to metres, and ground plane offset puts the ground at y = 0. The same page says the files use an OpenCV-style frame and that Marble's own viewer applies a half turn around the x axis. For our loft the numbers are 2.378 and 1.389 m. For the tavern, 2.237 and 1.577 m. The offset is the height of the point the world was generated from, above its floor. In lib/world.js the whole conversion is one transform, applied to the group that holds the splat: Scale, flip, lift. Because the scale is uniform and sits on the parent group, it scales the positions of the Gaussians and their sizes together, which is what World Labs' note asks for. The collider is in the same raw frame and gets the same matrix, so the two always agree. After this, [0, 0, 0] is the spot on the floor under the point the world was generated from, looking along -z. Everything the agent places is in plain metres.","u":"/blog/how-world-templates-work#from-raw-units-to-metres"},{"k":"blog","t":"How our world templates work","h":"The collider as a physics mesh","x":"A splat has no surfaces. The collider is a coarse triangle mesh of the same place; World Labs' export specs describe it as optimised for simple physics, with 100 to 200 thousand triangles. Ours have 102,206 and 217,939. <World loads it, moves every vertex into scene metres, and hands the result to Rapier as one triangle mesh on a fixed body: The FIX INTERNAL EDGES flag is there so that things roll","s":"A splat has no surfaces. The collider is a coarse triangle mesh of the same place; World Labs' export specs describe it as optimised for simple physics, with 100 to 200 thousand triangles. Ours have 102,206 and 217,939. <World loads it, moves every vertex into scene metres, and hands the result to Rapier as one triangle mesh on a fixed body: The FIX INTERNAL EDGES flag is there so that things roll and slide over the seams between triangles without bumps. The mesh is never drawn, except as a wireframe when you pass showCollider to see what is solid.","u":"/blog/how-world-templates-work#the-collider-as-a-physics-mesh"},{"k":"blog","t":"How our world templates work","h":"Flattening a lumpy floor","x":"The first thing we saw with the raw collider: its floor is not flat. In the loft it was up to 0.3 m above the floor you see. Anything standing on it would float above the visible floor. flattenFloor in lib/world.js presses it flat. The idea, in three steps: 1. Look only at triangles that lie within 0.4 m of y = 0 and face upwards. 2. Join those triangles into connected patches. A patch counts as f","s":"The first thing we saw with the raw collider: its floor is not flat. In the loft it was up to 0.3 m above the floor you see. Anything standing on it would float above the visible floor. flattenFloor in lib/world.js presses it flat. The idea, in three steps: 1. Look only at triangles that lie within 0.4 m of y = 0 and face upwards. 2. Join those triangles into connected patches. A patch counts as floor if it contains triangles that are nearly level and within 6 cm of y = 0, where the visible floor is. 3. Move every vertex of a floor patch to y = 0. Furniture, walls and anything reached only over a steep face keep their shape, because a steep triangle does not join patches. A short clean-up pass catches the few low vertices that sliver triangles leave standing. It runs once at load and took 11 to 20 ms for our two worlds on a laptop. It is on by default. For a world whose ground really slopes, such as hills, <World flatFloor={false} leaves the collider as it came.","u":"/blog/how-world-templates-work#flattening-a-lumpy-floor"},{"k":"blog","t":"How our world templates work","h":"A collider made from a single viewpoint","x":"The second limit cannot be pressed flat. The collider only knows what could be seen from the point the world was generated from. Space that was hidden from there is filled in solid. Behind the loft's worktable and behind the tavern's bar, the collider simply continues at table and bar height. The panorama of the tavern world. What cannot be seen from this point has no reliable shape in the collide","s":"The second limit cannot be pressed flat. The collider only knows what could be seen from the point the world was generated from. Space that was hidden from there is filled in solid. Behind the loft's worktable and behind the tavern's bar, the collider simply continues at table and bar height. The panorama of the tavern world. What cannot be seen from this point has no reliable shape in the collider. We did not try to repair this. We designed around it: - The game happens within sight of [0, 0, 0]. In World Game you stand in the aisle and throw at the bar; nothing needs to fall behind it. - World Explorer has a range that fences the walk to the part of the world that looks good and collides well. - Surfaces are a few centimetres off, so a row of mugs does not stand straight on the raw bar top. The Ledge component finds the highest point of the collider along a line and lays a level, invisible board there. The mugs stand on the board.","u":"/blog/how-world-templates-work#a-collider-made-from-a-single-viewpoint"},{"k":"blog","t":"How our world templates work","h":"Lighting added meshes from the splat","x":"Lights do nothing to a splat. Its light is baked in. But a red ball with flat lighting in a warm tavern looks pasted on. WorldLight asks Spark to render the world into an environment map from a point you choose, once, after the splat has loaded: Meshes with a standard material then pick up the colours of the place. The templates add one soft directional light for shape. Shadows are faked. A splat ","s":"Lights do nothing to a splat. Its light is baked in. But a red ball with flat lighting in a warm tavern looks pasted on. WorldLight asks Spark to render the world into an environment map from a point you choose, once, after the splat has loaded: Meshes with a standard material then pick up the colours of the place. The templates add one soft directional light for shape. Shadows are faked. A splat cannot receive one, so Placed puts a soft dark blob under what it places, and each ball in the game drops a blob onto whatever the collider says is below it.","u":"/blog/how-world-templates-work#lighting-added-meshes-from-the-splat"},{"k":"blog","t":"How our world templates work","h":"Walking","x":"World Explorer's Player is a capsule, 1.7 m tall with a 0.28 m radius, on a kinematic body. Rapier's character controller moves it: Each physics step, the template turns input into a wished movement and asks the controller how much of it the world allows. The capsule slides along walls, climbs ledges up to 0.3 m and follows the ground downhill. In the headset the left thumbstick walks where the he","s":"World Explorer's Player is a capsule, 1.7 m tall with a 0.28 m radius, on a kinematic body. Rapier's character controller moves it: Each physics step, the template turns input into a wished movement and asks the controller how much of it the world allows. The capsule slides along walls, climbs ledges up to 0.3 m and follows the ground downhill. In the headset the left thumbstick walks where the head looks, and the right one turns in 45° steps and teleports. Real steps in your room move the capsule too, so the world pushes back at walls. World Explorer on a flat screen: walk with W A S D, or take the guided tour.","u":"/blog/how-world-templates-work#walking"},{"k":"blog","t":"How our world templates work","h":"Throwing","x":"In World Game a ball is a dynamic body with a 5 cm radius and continuous collision detection on, so a fast throw does not pass through the thin shell of the collider. In the headset, Grabbable keeps a held ball dynamic. It is pulled to the hand by setting its velocity each physics step, so it pushes other things and cannot be carried through the world. On release it gets the hand's speed over the ","s":"In World Game a ball is a dynamic body with a 5 cm radius and continuous collision detection on, so a fast throw does not pass through the thin shell of the collider. In the headset, Grabbable keeps a held ball dynamic. It is pulled to the hand by setting its velocity each physics step, so it pushes other things and cannot be carried through the world. On release it gets the hand's speed over the last 80 ms, measured in real time so the throw does not depend on the frame rate. On a flat screen a drag looks around, so a click throws. The click's ray is cast into the physics world, and the ball is launched so that its arc passes through the point it hit: World Game in the emulated headset, both eyes. The mugs stand on an invisible level board laid on the bar.","u":"/blog/how-world-templates-work#throwing"},{"k":"blog","t":"How our world templates work","h":"How the agent switches worlds","x":"When you ask for a different place, the agent calls its generate world tool with a name and a description. The tool asks for your approval, waits the few minutes the generation takes, and saves the files into public/worlds/<name / . Then the agent changes one string: That is the whole switch for the world itself. What takes thought comes after it. Every world has its own furniture, so the agent re","s":"When you ask for a different place, the agent calls its generate world tool with a name and a description. The tool asks for your approval, waits the few minutes the generation takes, and saves the files into public/worlds/<name / . Then the agent changes one string: That is the whole switch for the world itself. What takes thought comes after it. Every world has its own furniture, so the agent reads the caption in world.json to learn what is there, uses heightAt or showCollider to find the surfaces, and moves what it had placed: the tour stops, the ledge, the targets, the scoreboard.","u":"/blog/how-world-templates-work#how-the-agent-switches-worlds"},{"k":"blog","t":"How our world templates work","h":"Read more","x":"- Graspable now generates worlds with World Labs: what it costs and how to try it - Spark: the renderer behind our splat templates - Performance of Gaussian splats in WebXR: what we measured and what we chose - Generated worlds, Templates and Credits in the documentation","u":"/blog/how-world-templates-work#read-more"},{"k":"blog","t":"How our world templates work","h":"Sources","x":"- Rendering Marble SPZ files in third-party engines, Export file specs and Interactive world examples, World Labs - SparkRenderer, Spark documentation - Colliders, Rigid bodies and Character controller, Rapier - react-three-rapier, pmndrs","u":"/blog/how-world-templates-work#sources"},{"k":"blog","t":"Spark: the renderer behind our splat templates","x":"What Spark is, why a Gaussian splat renderer that lives inside a normal three.js scene matters for a project an agent builds, what moving from Spark 0.1 to 2 changed, and how our templates set it up. Our three world templates show places as Gaussian splats. The thing that draws them is Spark, an open-source splat renderer for three.js built by World Labs, published as @sparkjsdev/spark . This arti","s":"What Spark is, why a Gaussian splat renderer that lives inside a normal three.js scene matters for a project an agent builds, what moving from Spark 0.1 to 2 changed, and how our templates set it up. Our three world templates show places as Gaussian splats. The thing that draws them is Spark, an open-source splat renderer for three.js built by World Labs, published as @sparkjsdev/spark . This article says why we use it, what changed when we moved from version 0.1 to 2, and how the templates set it up. If you want the background first, read Gaussian splats, explained.","u":"/blog/spark-splat-renderer"},{"k":"blog","t":"Spark: the renderer behind our splat templates","h":"What Spark is","x":"Spark draws 3D Gaussian splats with WebGL2 inside a three.js scene. Two classes matter: - SplatMesh is a splat object. It derives from THREE.Object3D , so it has a position, a rotation and a scale, sits anywhere in the scene graph, and loads from a file with new SplatMesh({ url }) . - SparkRenderer is one object you add to the scene. It collects the splats of every SplatMesh , sorts them for the c","s":"Spark draws 3D Gaussian splats with WebGL2 inside a three.js scene. Two classes matter: - SplatMesh is a splat object. It derives from THREE.Object3D , so it has a position, a rotation and a scale, sits anywhere in the scene graph, and loads from a file with new SplatMesh({ url }) . - SparkRenderer is one object you add to the scene. It collects the splats of every SplatMesh , sorts them for the current view and draws them. Spark's system design page describes the cycle. The renderer walks the visible scene graph and gathers all splats. Their distances from the viewpoint are read back from the GPU and sorted in a background worker. On the next render() call, all splats are drawn back to front in a single instanced draw call, and merged with opaque three.js geometry through the depth buffer.","u":"/blog/spark-splat-renderer#what-spark-is"},{"k":"blog","t":"Spark: the renderer behind our splat templates","h":"Why \"inside a normal three.js scene\" matters","x":"A Graspable project is a React Three Fiber project that an agent edits. The agent knows how to write meshes, materials, pointer handlers, physics bodies and @react-three/xr sessions. A splat renderer that needed its own canvas, its own camera or its own render loop would put the world outside everything the agent knows how to do. With Spark the world is one more object in the scene: - Splats and m","s":"A Graspable project is a React Three Fiber project that an agent edits. The agent knows how to write meshes, materials, pointer handlers, physics bodies and @react-three/xr sessions. A splat renderer that needed its own canvas, its own camera or its own render loop would put the world outside everything the agent knows how to do. With Spark the world is one more object in the scene: - Splats and meshes in one picture. Opaque meshes write depth, and splats are tested against it. A vase on a table is hidden by a plant in front of it and hides the wall behind it, with no extra work. - React Three Fiber. A SplatMesh goes into the scene like any three.js object, so the world can be a component with children. - WebXR. Spark renders through the same WebGLRenderer that three.js uses for a headset session. @react-three/xr starts the session as usual, and the splats are there in both eyes. Splat World on a flat screen. The room is one SplatMesh. The ceramic pieces and labels are ordinary meshes inside the same scene. There is one thing to remember when you add your own objects. Opaque meshes need nothing. Transparent ones, such as labels, glows and glass, need a renderOrder of 10 or higher and depthWrite={false} in our templates, or the splats are painted over them.","u":"/blog/spark-splat-renderer#why-inside-a-normal-three-js-scene-matters"},{"k":"blog","t":"Spark: the renderer behind our splat templates","h":"What moving from 0.1 to 2 changed for us","x":"The earlier Splat World template used Spark 0.1. The three world templates now use 2.3.1. Spark's migration guide lists everything; these three changes are the ones that shaped our code. A newer three.js. The guide names three.js r179 as the minimum, and the package declares three = 0.180.0 as a peer dependency. Our world templates pin three 0.180.0. Our other templates are on 0.170.0. The rendere","s":"The earlier Splat World template used Spark 0.1. The three world templates now use 2.3.1. Spark's migration guide lists everything; these three changes are the ones that shaped our code. A newer three.js. The guide names three.js r179 as the minimum, and the package declares three = 0.180.0 as a peer dependency. Our world templates pin three 0.180.0. Our other templates are on 0.170.0. The renderer is an object you create. In 0.1, Spark put a SparkRenderer into the scene for you. In 2 you create it and add it, or nothing is drawn: This looks like a chore. It turned out to be useful, because the renderer's settings now have one obvious home, and because the rule for the agent is simple to state: exactly one. With none, no splat is visible. With two, every splat is sorted and drawn twice. Level of detail. This is the main feature of Spark 2. Any splat file can be turned into a tree in which the original splats are the leaves and each level above is a coarser version. Every frame, Spark picks the best set of splats from the tree for the current viewpoint, within a fixed budget. The level-of-detail page gives the default budgets: 500,000 splats in a Quest, 750,000 in a Vision Pro, 1 million on Android, 1.5 million on iOS and 2.5 million on a desktop computer. lodSplatScale multiplies the budget. Turning it on is one flag, new SplatMesh({ url, lod: true }) . The tree is then built in a background worker when the file loads; the docs give 1 to 3 seconds per million input splats. Spark can also build the tree ahead of time and stream it from a .rad file. We do not use that yet: our worlds are 500,000 splats and are read from the project's own disk. Two smaller changes: sorting options were reduced to one switch, sortRadial , which is on by default and sorts by distance from the viewpoint so the order stays stable when the head turns. And Spark's VRButton was replaced by its own SparkXr helper. We use neither, because @react-three/xr already owns the session.","u":"/blog/spark-splat-renderer#what-moving-from-0-1-to-2-changed-for-us"},{"k":"blog","t":"Spark: the renderer behind our splat templates","h":"How the templates set it up","x":"Three small files, the same in all three templates. lib/spark.js holds every setting in one place (comments shortened here): components/Spark.jsx creates the one renderer and hands it to React Three Fiber as a primitive : maxStdDev is switched per frame because Spark's performance guide recommends Math.sqrt(5) for VR and the default Math.sqrt(8) otherwise. components/World.jsx creates the splat. I","s":"Three small files, the same in all three templates. lib/spark.js holds every setting in one place (comments shortened here): components/Spark.jsx creates the one renderer and hands it to React Three Fiber as a primitive : maxStdDev is switched per frame because Spark's performance guide recommends Math.sqrt(5) for VR and the default Math.sqrt(8) otherwise. components/World.jsx creates the splat. It is made once, in an effect, never in render and never per frame. The core of it: The small 100,000-splat file is shown at once. When the 500,000-splat file has loaded and its level-of-detail tree is built, the two are swapped. Pointer rays never test the splat ( raycastable: false ): what is solid is decided by the collider mesh, which is the subject of How our world templates work. In App.jsx it all comes together:","u":"/blog/spark-splat-renderer#how-the-templates-set-it-up"},{"k":"blog","t":"Spark: the renderer behind our splat templates","h":"One more use: light from the splat","x":"A splat cannot be lit, but it can light other things. SparkRenderer.renderEnvMap renders the splats around a point into a cube map and filters it for image-based lighting. Our WorldLight component calls it once after the world has loaded and sets the result as scene.environment . Meshes with a standard material then take on the colours of the place. The tavern world in the emulated headset, both e","s":"A splat cannot be lit, but it can light other things. SparkRenderer.renderEnvMap renders the splats around a point into a cube map and filters it for image-based lighting. Our WorldLight component calls it once after the world has loaded and sets the result as scene.environment . Meshes with a standard material then take on the colours of the place. The tavern world in the emulated headset, both eyes. Splats, meshes and the score panel line up in each eye.","u":"/blog/spark-splat-renderer#one-more-use-light-from-the-splat"},{"k":"blog","t":"Spark: the renderer behind our splat templates","h":"What we would tell someone starting out","x":"- Add exactly one SparkRenderer , and create the WebGLRenderer with antialias: false . - Create each SplatMesh once. Do not clone a world to \"duplicate\" it. - Turn level of detail on and leave the budget to Spark's per-device default until you have measured on the device. - Do not add post-processing passes. The splats are already the expensive part. The numbers behind those choices are in Perform","s":"- Add exactly one SparkRenderer , and create the WebGLRenderer with antialias: false . - Create each SplatMesh once. Do not clone a world to \"duplicate\" it. - Turn level of detail on and leave the budget to Spark's per-device default until you have measured on the device. - Do not add post-processing passes. The splats are already the expensive part. The numbers behind those choices are in Performance of Gaussian splats in WebXR: what we measured and what we chose. To try the templates, see Templates and Generated worlds; the announcement is here.","u":"/blog/spark-splat-renderer#what-we-would-tell-someone-starting-out"},{"k":"blog","t":"Spark: the renderer behind our splat templates","h":"Sources","x":"- Spark and its source repository, World Labs - System design, New features in 2.0, 0.1 → 2.0 migration guide, Getting started with level of detail, SparkRenderer, SplatMesh and Performance tuning, Spark documentation - Interactive world examples, World Labs","u":"/blog/spark-splat-renderer#sources"},{"k":"blog","t":"How to test a WebXR app without a headset","x":"Emulate a headset in the browser with IWER: install it, move the headset and controllers from code, use a synthetic room for AR, record a session and replay it. Putting a headset on for every small change is slow. Most of the time you only need to know that the scene starts, that the trigger does what it should, and that nothing broke. An emulated headset in a desktop browser answers those questio","s":"Emulate a headset in the browser with IWER: install it, move the headset and controllers from code, use a synthetic room for AR, record a session and replay it. Putting a headset on for every small change is slow. Most of the time you only need to know that the scene starts, that the trigger does what it should, and that nothing broke. An emulated headset in a desktop browser answers those questions in seconds. This guide shows how, and says clearly what emulation cannot tell you.","u":"/blog/test-webxr-without-a-headset"},{"k":"blog","t":"How to test a WebXR app without a headset","h":"The short answer","x":"Use IWER or Meta's Immersive Web Emulator to exercise a WebXR session, including headset poses and controller or hand input. Record a representative interaction and replay it after changes, checking the resulting scene. Use the physical headset for tracking, comfort and performance checks. Desktop orbit controls help inspect a scene, but do not test XR input. Chrome's Sensors panel can simulate or","s":"Use IWER or Meta's Immersive Web Emulator to exercise a WebXR session, including headset poses and controller or hand input. Record a representative interaction and replay it after changes, checking the resulting scene. Use the physical headset for tracking, comfort and performance checks. Desktop orbit controls help inspect a scene, but do not test XR input. Chrome's Sensors panel can simulate orientation and touch; that is different from creating an immersive session with tracked controllers and hit-test surfaces.","u":"/blog/test-webxr-without-a-headset#the-short-answer"},{"k":"blog","t":"How to test a WebXR app without a headset","h":"Three ways to emulate","x":"- IWER in the page. The Immersive Web Emulation Runtime is a JavaScript library from Meta. It installs a WebXR implementation into the page, so navigator.xr exists in a browser that has none. You control the emulated device from code. - A browser extension. The Immersive Web Emulator is an extension for Chromium-based browsers that adds an emulated headset to any page. It needs no change to your p","s":"- IWER in the page. The Immersive Web Emulation Runtime is a JavaScript library from Meta. It installs a WebXR implementation into the page, so navigator.xr exists in a browser that has none. You control the emulated device from code. - A browser extension. The Immersive Web Emulator is an extension for Chromium-based browsers that adds an emulated headset to any page. It needs no change to your project, and it is built on IWER. - The emulator inside @react-three/xr. createXRStore has an emulate option. By default it emulates a Meta Quest 3 when the page is on localhost and the browser has no WebXR. It also uses IWER. The extension is the quickest way to look around by hand. IWER in the page is the one to choose when you want a script, or a test, to drive the headset.","u":"/blog/test-webxr-without-a-headset#three-ways-to-emulate"},{"k":"blog","t":"How to test a WebXR app without a headset","h":"Installing IWER in the page","x":"Create the device and install the runtime before your app asks for a session. Make desktop emulation an explicit development choice. In this standalone Vite example, add ?emulateXR=1 to the desktop preview address: Leave that parameter off when opening the development page on a physical headset. A development server can be visited by real hardware too; checking only import.meta.env.DEV does not di","s":"Create the device and install the runtime before your app asks for a session. Make desktop emulation an explicit development choice. In this standalone Vite example, add ?emulateXR=1 to the desktop preview address: Leave that parameter off when opening the development page on a physical headset. A development server can be visited by real hardware too; checking only import.meta.env.DEV does not distinguish a laptop from a Quest. IWER's runtime guidance explains that installing its runtime can override native WebXR. Vite removes this branch from a production build, so the emulator is excluded when this is its only import. Graspable manages its preview runtime itself; this example is for your own Vite project. If you use @react-three/xr, switch its own emulator off so that two emulators do not compete:","u":"/blog/test-webxr-without-a-headset#installing-iwer-in-the-page"},{"k":"blog","t":"How to test a WebXR app without a headset","h":"Driving the headset from code","x":"Once the runtime is installed, the device is an object you can move. Positions are in metres. Hands work the same way: These calls are described in the IWER getting started guide. The release above waits a frame so that the app sees the press before it sees the release.","u":"/blog/test-webxr-without-a-headset#driving-the-headset-from-code"},{"k":"blog","t":"How to test a WebXR app without a headset","h":"On-screen controls and a room for AR","x":"Two add-on packages make the emulator usable by hand and for AR: @iwer/devui adds on-screen controls for the headset, controllers and hands. While they are installed they set the device's state on every frame, so drive the device either from the controls or from your own code, not both at once. @iwer/sem puts the emulated headset in a synthetic room with floors, walls and furniture. That matters f","s":"Two add-on packages make the emulator usable by hand and for AR: @iwer/devui adds on-screen controls for the headset, controllers and hands. While they are installed they set the device's state on every frame, so drive the device either from the controls or from your own code, not both at once. @iwer/sem puts the emulated headset in a synthetic room with floors, walls and furniture. That matters for AR: without a room there are no surfaces, so hit tests return nothing and plane detection finds no planes. Use those surfaces to test placement logic; they do not reproduce tracking noise or the geometry of the actual room.","u":"/blog/test-webxr-without-a-headset#on-screen-controls-and-a-room-for-ar"},{"k":"blog","t":"How to test a WebXR app without a headset","h":"Record a session and replay it","x":"Scripting every movement by hand gets tedious. IWER can record what you do in a session and play it back. Play it back with the device: The player uses the recorded timestamps, so a replay on a faster or slower machine still follows the same movements. See action recording and playback in the IWER documentation. A recording becomes a test once you compare something after the replay. Useful things ","s":"Scripting every movement by hand gets tedious. IWER can record what you do in a session and play it back. Play it back with the device: The player uses the recorded timestamps, so a replay on a faster or slower machine still follows the same movements. See action recording and playback in the IWER documentation. A recording becomes a test once you compare something after the replay. Useful things to compare: - Errors in the console. There should be none. - The scene: which objects exist, and where they are. - A screenshot of the last frame. Screenshots change with animation and with the graphics driver, so treat a difference as a hint and not as a failure. Because all of this runs in an ordinary browser, it also runs in a browser without a window, on your computer or in continuous integration.","u":"/blog/test-webxr-without-a-headset#record-a-session-and-replay-it"},{"k":"blog","t":"How to test a WebXR app without a headset","h":"What emulation does not tell you","x":"Be honest with yourself about this list. An emulated headset is a model of the device, not the device. - Performance. Your computer's graphics card is not a headset's mobile chip. Frame rate has to be measured on the headset. - Comfort. Scale, distance, reach and motion sickness are felt, not computed. - Real tracking. Emulated hands are clean poses. Real hands are noisy, and tracking is sometimes","s":"Be honest with yourself about this list. An emulated headset is a model of the device, not the device. - Performance. Your computer's graphics card is not a headset's mobile chip. Frame rate has to be measured on the headset. - Comfort. Scale, distance, reach and motion sickness are felt, not computed. - Real tracking. Emulated hands are clean poses. Real hands are noisy, and tracking is sometimes lost. - The real room. A synthetic room is tidy. A real one has glass, clutter and bad light. - Browser differences. The emulator implements the standard. Each headset's browser has its own gaps and quirks. A sensible routine is to use the emulator for every change, and a headset before you call something finished.","u":"/blog/test-webxr-without-a-headset#what-emulation-does-not-tell-you"},{"k":"blog","t":"How to test a WebXR app without a headset","h":"Doing this in Graspable","x":"Graspable sets this up for web projects. The desktop preview includes an emulated headset, with on-screen controls and a choice of synthetic rooms for AR. There is nothing to install or configure. A game in the emulated headset, standing in a synthetic living room. The preview has Record and replay buttons. A recording is saved in the project's tests/xr/ folder, and after each of the agent's runs ","s":"Graspable sets this up for web projects. The desktop preview includes an emulated headset, with on-screen controls and a choice of synthetic rooms for AR. There is nothing to install or configure. A game in the emulated headset, standing in a synthetic living room. The preview has Record and replay buttons. A recording is saved in the project's tests/xr/ folder, and after each of the agent's runs Graspable replays every recording in a browser without a window and compares the scene. If something you recorded stops working, you hear about it before you put the headset on. See Preview and headset and XR tests, or read how Graspable checks its own work. When you are ready for the real device, the preview can open the project on a headset over Wi-Fi or USB.","u":"/blog/test-webxr-without-a-headset#doing-this-in-graspable"},{"k":"blog","t":"How to test a WebXR app without a headset","h":"Sources","x":"- Sensors panel, Chrome DevTools - Immersive Web Emulation Runtime: getting started, action recording and playback, Meta - Immersive Web Emulator, Meta - WebXR Device API, W3C - Store options, @react-three/xr","u":"/blog/test-webxr-without-a-headset#sources"},{"k":"blog","t":"Hand tracking or controllers in WebXR, and how to support both","x":"How WebXR describes hands and controllers, what each can and cannot do, and how to write one React Three Fiber scene that works with either. A headset user can pick up the controllers or put them down at any moment. A WebXR app that handles only one of the two stops working halfway through a session. This guide explains how WebXR sees hands and controllers, where they differ, and how to support bo","s":"How WebXR describes hands and controllers, what each can and cannot do, and how to write one React Three Fiber scene that works with either. A headset user can pick up the controllers or put them down at any moment. A WebXR app that handles only one of the two stops working halfway through a session. This guide explains how WebXR sees hands and controllers, where they differ, and how to support both with @react-three/xr.","u":"/blog/webxr-hand-tracking-vs-controllers"},{"k":"blog","t":"Hand tracking or controllers in WebXR, and how to support both","h":"One model for every input","x":"WebXR describes every way of pointing as an XRInputSource . A controller is one. A tracked hand is one. A tap on a phone screen is one. Each has: - handedness : 'left' , 'right' or 'none' . - targetRaySpace : where it points. Use this for aiming. - gripSpace : where it is held. Use this for showing a model or holding an object. - targetRayMode : how it aims, for example 'tracked-pointer' for a con","s":"WebXR describes every way of pointing as an XRInputSource . A controller is one. A tracked hand is one. A tap on a phone screen is one. Each has: - handedness : 'left' , 'right' or 'none' . - targetRaySpace : where it points. Use this for aiming. - gripSpace : where it is held. Use this for showing a model or holding an object. - targetRayMode : how it aims, for example 'tracked-pointer' for a controller or hand. - profiles : names that identify the hardware, listed in the WebXR input profiles registry. Two properties tell hands and controllers apart: - A controller has a gamepad with buttons and thumbstick axes, defined by the WebXR Gamepads Module. - A tracked hand has a hand : an XRHand with 25 joints, defined by the WebXR Hand Input Module. It is only there if the session was started with the hand-tracking feature. Both fire the same events. select is the main action: a trigger press on a controller, a pinch of thumb and index finger with hands on Meta Quest. squeeze is the grip button. If your app listens for select and aims with targetRaySpace , it already works with both.","u":"/blog/webxr-hand-tracking-vs-controllers#one-model-for-every-input"},{"k":"blog","t":"Hand tracking or controllers in WebXR, and how to support both","h":"What differs","x":"Controllers have: - Buttons, a trigger, a grip and a thumbstick, each with an exact value. - Haptics. The controller can vibrate. - Motion sensors inside, so tracking can carry on for a moment when the cameras cannot see the controller. Hands have: - Nothing to pick up. People can start at once. - 25 joint poses per hand, so you can detect poses and touch things with a fingertip. - No buttons and ","s":"Controllers have: - Buttons, a trigger, a grip and a thumbstick, each with an exact value. - Haptics. The controller can vibrate. - Motion sensors inside, so tracking can carry on for a moment when the cameras cannot see the controller. Hands have: - Nothing to pick up. People can start at once. - 25 joint poses per hand, so you can detect poses and touch things with a fingertip. - No buttons and no thumbstick. Moving around needs another design, such as teleporting by pinching. - No haptics. - Tracking that is lost when the cameras cannot see the hand, for example when one hand covers the other. Do not guess which one your users prefer. Support both, and if you need a number, measure it with your own app and your own users.","u":"/blog/webxr-hand-tracking-vs-controllers#what-differs"},{"k":"blog","t":"Hand tracking or controllers in WebXR, and how to support both","h":"Reacting when the input changes","x":"Input sources come and go during a session. The session fires inputsourceschange when a controller is put down and a hand appears in its place. Never store an input source at session start and keep using it. With @react-three/xr you do not handle that event yourself. The hooks below return the current state and re-render when it changes.","u":"/blog/webxr-hand-tracking-vs-controllers#reacting-when-the-input-changes"},{"k":"blog","t":"Hand tracking or controllers in WebXR, and how to support both","h":"One scene for both","x":"@react-three/xr gives hands and controllers the same pointers: a ray for things far away, a grab pointer for things within reach, and, for hands, a touch pointer on the index finger. They all send ordinary React Three Fiber pointer events. So the main interaction needs no branching: See the interactions tutorial for the full list of pointer events.","u":"/blog/webxr-hand-tracking-vs-controllers#one-scene-for-both"},{"k":"blog","t":"Hand tracking or controllers in WebXR, and how to support both","h":"Reading a thumbstick","x":"When you need something only a controller has, ask for the controller's state and handle its absence. useXRInputSourceState returns undefined when that controller is not there, for example because the user switched to hands. The names such as xr-standard-thumbstick and xr-standard-trigger come from the input profiles registry. The gamepad tutorial lists them.","u":"/blog/webxr-hand-tracking-vs-controllers#reading-a-thumbstick"},{"k":"blog","t":"Hand tracking or controllers in WebXR, and how to support both","h":"Reading a hand joint","x":"When you need something only a hand has, ask for the hand's state. This puts a small sphere on the tip of the right index finger: XRSpace keeps its children at the pose of that joint on every frame. In the plain API the same thing is frame.getJointPose(joint, referenceSpace) , which also returns the joint's radius.","u":"/blog/webxr-hand-tracking-vs-controllers#reading-a-hand-joint"},{"k":"blog","t":"Hand tracking or controllers in WebXR, and how to support both","h":"Listening for a pinch or a trigger","x":"For the main action, listen for select and you get both: For your own gestures, such as a fist or a thumbs up, compare joint positions yourself. Pick the threshold by trying it on a headset with several people. Hands differ in size.","u":"/blog/webxr-hand-tracking-vs-controllers#listening-for-a-pinch-or-a-trigger"},{"k":"blog","t":"Hand tracking or controllers in WebXR, and how to support both","h":"A short checklist","x":"- Aim with targetRaySpace and act on select . That covers both. - Make every target reachable by a ray, because a hand has no thumbstick to move closer with. - Offer a way to move that needs no thumbstick. - Treat controller state and hand state as things that can be undefined on any frame. - Do not depend on haptics to tell the user something happened. Show it or play a sound as well. - Test the ","s":"- Aim with targetRaySpace and act on select . That covers both. - Make every target reachable by a ray, because a hand has no thumbstick to move closer with. - Offer a way to move that needs no thumbstick. - Treat controller state and hand state as things that can be undefined on any frame. - Do not depend on haptics to tell the user something happened. Show it or play a sound as well. - Test the switch: start with controllers, put them down, carry on with hands.","u":"/blog/webxr-hand-tracking-vs-controllers#a-short-checklist"},{"k":"blog","t":"Hand tracking or controllers in WebXR, and how to support both","h":"Doing this in Graspable","x":"Projects made in Graspable use @react-three/xr, so the code in this guide fits them as it is. You can ask the agent for the behaviour instead, for example: \"Make the button work with hand tracking as well as controllers, and add teleport by pinching.\" The emulated headset in the preview has both controllers and hands, so you can try both kinds of input before you put a headset on. See Preview and ","s":"Projects made in Graspable use @react-three/xr, so the code in this guide fits them as it is. You can ask the agent for the behaviour instead, for example: \"Make the button work with hand tracking as well as controllers, and add teleport by pinching.\" The emulated headset in the preview has both controllers and hands, so you can try both kinds of input before you put a headset on. See Preview and headset. Emulated hands are simple poses. How well a gesture works with real fingers is something to test on a real headset.","u":"/blog/webxr-hand-tracking-vs-controllers#doing-this-in-graspable"},{"k":"blog","t":"Hand tracking or controllers in WebXR, and how to support both","h":"Sources","x":"- XRInputSource and XRHand, MDN - WebXR Hand Input Module, W3C - WebXR Gamepads Module, W3C - WebXR input profiles registry, Immersive Web Working Group - Interactions, custom inputs and gamepad tutorials, @react-three/xr","u":"/blog/webxr-hand-tracking-vs-controllers#sources"},{"k":"blog","t":"WebXR hit testing: how to place objects on real surfaces","x":"What a WebXR hit test is, how to request one, how to read its result, and a working reticle in React Three Fiber that places objects on floors and tables. To put a virtual chair on a real floor, the app has to ask the device where the floor is. In WebXR that question is a hit test. This guide explains what it is, shows the plain API, and then a version in React Three Fiber that you can paste into ","s":"What a WebXR hit test is, how to request one, how to read its result, and a working reticle in React Three Fiber that places objects on floors and tables. To put a virtual chair on a real floor, the app has to ask the device where the floor is. In WebXR that question is a hit test. This guide explains what it is, shows the plain API, and then a version in React Three Fiber that you can paste into a project.","u":"/blog/webxr-hit-testing-place-objects-on-real-surfaces"},{"k":"blog","t":"WebXR hit testing: how to place objects on real surfaces","h":"What a hit test is","x":"A hit test casts a ray into the device's understanding of the real world and returns the places where the ray meets a real surface. It is defined by the WebXR Hit Test Module. It is not the raycasting you know from three.js. A three.js Raycaster tests your own meshes. A WebXR hit test knows nothing about your meshes. It tests against what the device has detected: planes, a room mesh, or depth poin","s":"A hit test casts a ray into the device's understanding of the real world and returns the places where the ray meets a real surface. It is defined by the WebXR Hit Test Module. It is not the raycasting you know from three.js. A three.js Raycaster tests your own meshes. A WebXR hit test knows nothing about your meshes. It tests against what the device has detected: planes, a room mesh, or depth points, depending on the device. Each result gives you a pose, which is a position and an orientation. The specification defines the orientation so that the pose's Y axis points along the surface normal. On a floor, Y points up. On a wall, Y points out of the wall. That is what lets you lay a reticle flat on any surface.","u":"/blog/webxr-hit-testing-place-objects-on-real-surfaces#what-a-hit-test-is"},{"k":"blog","t":"WebXR hit testing: how to place objects on real surfaces","h":"What you need","x":"- An immersive-ar session. Hit testing is an AR feature. - The hit-test feature, requested when the session starts. Without it, requestHitTestSource throws. - A browser that implements it. Chrome on Android and the Meta Quest Browser do. At the time of writing, Safari on iPhone has no WebXR. Ask the browser with isSessionSupported and do not assume. - A secure page. WebXR works over HTTPS and on l","s":"- An immersive-ar session. Hit testing is an AR feature. - The hit-test feature, requested when the session starts. Without it, requestHitTestSource throws. - A browser that implements it. Chrome on Android and the Meta Quest Browser do. At the time of writing, Safari on iPhone has no WebXR. Ask the browser with isSessionSupported and do not assume. - A secure page. WebXR works over HTTPS and on localhost .","u":"/blog/webxr-hit-testing-place-objects-on-real-surfaces#what-you-need"},{"k":"blog","t":"WebXR hit testing: how to place objects on real surfaces","h":"The plain API","x":"The steps are: start a session, create a hit test source once, then read its results on every frame. Three things in this code matter: - The source is created once. requestHitTestSource returns a promise. Create the source after the session starts and keep it. Do not create one per frame. - Results are read per frame. getHitTestResults is synchronous and only valid inside the frame callback. - The","s":"The steps are: start a session, create a hit test source once, then read its results on every frame. Three things in this code matter: - The source is created once. requestHitTestSource returns a promise. Create the source after the session starts and keep it. Do not create one per frame. - Results are read per frame. getHitTestResults is synchronous and only valid inside the frame callback. - The space decides where the ray comes from. With viewer , the ray leaves the centre of the view. That suits a phone, where you aim with the screen. On a headset you usually aim with a controller or a hand, so you pass that input source's targetRaySpace instead. You can also say what to test against with entityTypes : 'plane' , 'point' or 'mesh' . The default is planes only.","u":"/blog/webxr-hit-testing-place-objects-on-real-surfaces#the-plain-api"},{"k":"blog","t":"WebXR hit testing: how to place objects on real surfaces","h":"A reticle in React Three Fiber","x":"@react-three/xr wraps this in a hook. createXRStore requests the hit-test feature by default, so there is nothing to switch on. This component shows a ring where the ray meets a surface, and places a small box there when you press the trigger, pinch, or tap the screen. getWorldMatrix writes the hit pose into a three.js matrix in world coordinates and returns false if the pose is not available this","s":"@react-three/xr wraps this in a hook. createXRStore requests the hit-test feature by default, so there is nothing to switch on. This component shows a ring where the ray meets a surface, and places a small box there when you press the trigger, pinch, or tap the screen. getWorldMatrix writes the hit pose into a three.js matrix in world coordinates and returns false if the pose is not available this frame. Decomposing that matrix into the group gives the reticle both its position and its tilt.","u":"/blog/webxr-hit-testing-place-objects-on-real-surfaces#a-reticle-in-react-three-fiber"},{"k":"blog","t":"WebXR hit testing: how to place objects on real surfaces","h":"Aiming with a controller instead of the head","x":"On a headset, a ray from the head is awkward. Give each controller its own hit test with the XRHitTest component, placed inside a custom controller: The same works for hands with the hand option.","u":"/blog/webxr-hit-testing-place-objects-on-real-surfaces#aiming-with-a-controller-instead-of-the-head"},{"k":"blog","t":"WebXR hit testing: how to place objects on real surfaces","h":"Keeping an object where you put it","x":"A hit result is true for one frame. As the device learns more about the room, its coordinate system shifts slightly, and an object placed at a fixed position can drift away from the real spot. An anchor fixes that. The device keeps an anchor attached to the real place and updates its pose for you. You can create one straight from a hit result with XRHitTestResult.createAnchor() , if the session ha","s":"A hit result is true for one frame. As the device learns more about the room, its coordinate system shifts slightly, and an object placed at a fixed position can drift away from the real spot. An anchor fixes that. The device keeps an anchor attached to the real place and updates its pose for you. You can create one straight from a hit result with XRHitTestResult.createAnchor() , if the session has the anchors feature. See the WebXR Anchors Module and the anchors tutorial for @react-three/xr. For a short session in one room, a plain position is often good enough. For anything that must stay put for minutes, use an anchor.","u":"/blog/webxr-hit-testing-place-objects-on-real-surfaces#keeping-an-object-where-you-put-it"},{"k":"blog","t":"WebXR hit testing: how to place objects on real surfaces","h":"When it does not work","x":"- No results at first. The device needs to see the surface. On a phone, move it slowly from side to side. On a headset, the room may need to be set up first. - requestHitTestSource throws. The feature was not requested, or the device refused it. Ask for it as an optional feature and check session.enabledFeatures if your app can work without it. - The reticle stands upright on the floor. The pose i","s":"- No results at first. The device needs to see the surface. On a phone, move it slowly from side to side. On a headset, the room may need to be set up first. - requestHitTestSource throws. The feature was not requested, or the device refused it. Ask for it as an optional feature and check session.enabledFeatures if your app can work without it. - The reticle stands upright on the floor. The pose is right and your mesh is not. Ring and plane geometries face +Z, and the surface normal is +Y. - Results only on floors and tables. You are testing planes only. What else is available depends on the device. - It works on the phone and not on the headset. Check which space the ray starts from.","u":"/blog/webxr-hit-testing-place-objects-on-real-surfaces#when-it-does-not-work"},{"k":"blog","t":"WebXR hit testing: how to place objects on real surfaces","h":"Doing this in Graspable","x":"In Graspable you can ask for this in a sentence, for example: \"Build an AR measuring tape. I point at a surface, pull the trigger to place points, and see the length in centimetres.\" The agent writes code like the above into the project on your computer. The useful part for hit testing is the preview. Its emulated headset stands in a synthetic room, so hit tests return real results and you can pla","s":"In Graspable you can ask for this in a sentence, for example: \"Build an AR measuring tape. I point at a surface, pull the trigger to place points, and see the length in centimetres.\" The agent writes code like the above into the project on your computer. The useful part for hit testing is the preview. Its emulated headset stands in a synthetic room, so hit tests return real results and you can place things on the floor without putting a headset on. You can record that session and Graspable replays it after later changes. See Preview and headset and XR tests. An emulated room is not your room. Before you rely on it, open the project on a real headset.","u":"/blog/webxr-hit-testing-place-objects-on-real-surfaces#doing-this-in-graspable"},{"k":"blog","t":"WebXR hit testing: how to place objects on real surfaces","h":"Sources","x":"- WebXR Hit Test Module, W3C - XRSession.requestHitTestSource(), XRFrame.getHitTestResults() and XRHitTestResult, MDN - WebXR Anchors Module, Immersive Web Working Group - Hit test tutorial, @react-three/xr","u":"/blog/webxr-hit-testing-place-objects-on-real-surfaces#sources"},{"k":"blog","t":"A performance budget for WebXR on Quest 3","x":"How much time a frame has in the Meta Quest Browser, how to find out whether the CPU or the GPU is the limit, and what to change in a React Three Fiber scene. A WebXR scene that runs smoothly on a laptop can stutter on a headset. The headset has a mobile chip, draws everything twice, and has to do it far more often per second than a film. A missed frame in a headset is not a cosmetic problem. The ","s":"How much time a frame has in the Meta Quest Browser, how to find out whether the CPU or the GPU is the limit, and what to change in a React Three Fiber scene. A WebXR scene that runs smoothly on a laptop can stutter on a headset. The headset has a mobile chip, draws everything twice, and has to do it far more often per second than a film. A missed frame in a headset is not a cosmetic problem. The world visibly judders, and people feel unwell. This guide gives you the one number that is fixed, the frame time, and a method to find the rest for your own scene. It does not give triangle or draw call limits, because honest ones depend on your materials, your lights and your code. You have to measure them.","u":"/blog/webxr-performance-budget-quest-3"},{"k":"blog","t":"A performance budget for WebXR on Quest 3","h":"The budget is time","x":"The headset shows a new image at a fixed rate. Your frame has to be finished before the next one is due. The time per frame is 1000 ms divided by the rate: - 72 frames per second: 13.9 ms - 90 frames per second: 11.1 ms - 120 frames per second: 8.3 ms Meta's documentation says a WebXR session in the Meta Quest Browser runs at 72 frames per second by default on Quest headsets, and that the app can ","s":"The headset shows a new image at a fixed rate. Your frame has to be finished before the next one is due. The time per frame is 1000 ms divided by the rate: - 72 frames per second: 13.9 ms - 90 frames per second: 11.1 ms - 120 frames per second: 8.3 ms Meta's documentation says a WebXR session in the Meta Quest Browser runs at 72 frames per second by default on Quest headsets, and that the app can ask for another rate. The rates a device offers are in session.supportedFrameRates , and session.updateTargetFrameRate(rate) asks for one. See WebXR app framerate control. That time is shared by everything: your JavaScript, React, physics, three.js preparing the scene, and the graphics chip drawing it for two eyes. The browser and the system need some of it too. So aim to finish well inside the budget, not at its edge. A higher rate is not always better. A scene that holds 72 without missing a frame is more comfortable than one that asks for 90 and misses often.","u":"/blog/webxr-performance-budget-quest-3#the-budget-is-time"},{"k":"blog","t":"A performance budget for WebXR on Quest 3","h":"Why a headset is harder than a laptop","x":"- Two views. Each eye gets its own image. three.js 0.170 draws the scene once per eye, so every draw call happens twice. - Many pixels. The two images together have far more pixels than a typical laptop window, and each pixel runs your fragment shader. - A mobile chip. It has less power and a tighter heat limit than a desktop graphics card. When it gets hot it slows down.","u":"/blog/webxr-performance-budget-quest-3#why-a-headset-is-harder-than-a-laptop"},{"k":"blog","t":"A performance budget for WebXR on Quest 3","h":"First find the limit: CPU or GPU","x":"Changing things at random wastes time. Find out which side is over budget first. Meta's performance workflow describes the method, and it comes down to one experiment: make the picture much smaller and see whether the frame rate recovers. - If it recovers, you are limited by pixels, which is the GPU drawing fragments. Reduce resolution, overdraw, shader cost and post-processing. - If it does not, ","s":"Changing things at random wastes time. Find out which side is over budget first. Meta's performance workflow describes the method, and it comes down to one experiment: make the picture much smaller and see whether the frame rate recovers. - If it recovers, you are limited by pixels, which is the GPU drawing fragments. Reduce resolution, overdraw, shader cost and post-processing. - If it does not, you are limited by the CPU or by geometry. Reduce draw calls, JavaScript work per frame, and vertex count. With @react-three/xr the experiment is one option: Tools for looking closer: - OVR Metrics Tool shows frame rate and CPU and GPU load inside the headset while your app runs. - Remote debugging. Connect the headset by USB and open chrome://inspect on your computer to use the Chrome developer tools, including the Performance panel, on the page in the headset. See debugging the browser remotely.","u":"/blog/webxr-performance-budget-quest-3#first-find-the-limit-cpu-or-gpu"},{"k":"blog","t":"A performance budget for WebXR on Quest 3","h":"Count what you draw","x":"three.js keeps counters for the last frame in renderer.info . Print them while the session runs: Put <RenderStats / inside the Canvas . In a session the numbers cover both eyes. This is how you get your own budget. Run the scene on the headset, watch the frame rate, and add content until frames start to drop. The counts just before that point are the budget for this scene, with these materials. Wr","s":"three.js keeps counters for the last frame in renderer.info . Print them while the session runs: Put <RenderStats / inside the Canvas . In a session the numbers cover both eyes. This is how you get your own budget. Run the scene on the headset, watch the frame rate, and add content until frames start to drop. The counts just before that point are the budget for this scene, with these materials. Write them down and check them when the scene grows.","u":"/blog/webxr-performance-budget-quest-3#count-what-you-draw"},{"k":"blog","t":"A performance budget for WebXR on Quest 3","h":"Reduce draw calls","x":"Each mesh with its own material is at least one draw call per eye. Many small objects cost more than one large one with the same number of triangles. - Instancing. Many copies of the same geometry and material can be one draw call with InstancedMesh . - Merging. Static objects that share a material can be merged into one geometry. - Sharing materials. Fewer distinct materials means fewer shader pr","s":"Each mesh with its own material is at least one draw call per eye. Many small objects cost more than one large one with the same number of triangles. - Instancing. Many copies of the same geometry and material can be one draw call with InstancedMesh . - Merging. Static objects that share a material can be merged into one geometry. - Sharing materials. Fewer distinct materials means fewer shader programs and state changes. A thousand boxes as one draw call per eye:","u":"/blog/webxr-performance-budget-quest-3#reduce-draw-calls"},{"k":"blog","t":"A performance budget for WebXR on Quest 3","h":"Reduce the cost per pixel","x":"- Use the cheapest material that looks right. meshBasicMaterial does no lighting. meshLambertMaterial is cheaper than meshStandardMaterial . Baked lighting in a texture costs nothing at run time. - Few lights, and be careful with shadows. Every shadow-casting light draws the scene again into a shadow map. - Avoid overdraw. Large transparent surfaces stacked in front of each other make the chip dra","s":"- Use the cheapest material that looks right. meshBasicMaterial does no lighting. meshLambertMaterial is cheaper than meshStandardMaterial . Baked lighting in a texture costs nothing at run time. - Few lights, and be careful with shadows. Every shadow-casting light draws the scene again into a shadow map. - Avoid overdraw. Large transparent surfaces stacked in front of each other make the chip draw the same pixel several times. - Be careful with post-processing. Each full-screen pass draws every pixel of both eyes again. It also works against foveation, which Meta's documentation says only applies when rendering straight to the final frame buffer. - Use foveation. Fixed foveated rendering draws the edges of each eye's image at a lower resolution, where the lens blurs them anyway. It is a value from 0 to 1 on the session's layer, XRWebGLLayer.fixedFoveation . See Meta's page on fixed foveated rendering. @react-three/xr exposes foveation, resolution and frame rate as store options: Its performance page describes them.","u":"/blog/webxr-performance-budget-quest-3#reduce-the-cost-per-pixel"},{"k":"blog","t":"A performance budget for WebXR on Quest 3","h":"Reduce JavaScript work per frame","x":"- Do not set React state every frame. Change objects directly inside useFrame . A state update re-renders components, and at 72 times a second that is a lot of work. See the React Three Fiber pitfalls. - Do not create objects every frame. new Vector3() inside useFrame produces garbage, and the garbage collector pauses the page to clean it up. Create helpers once, outside the loop, and reuse them. ","s":"- Do not set React state every frame. Change objects directly inside useFrame . A state update re-renders components, and at 72 times a second that is a lot of work. See the React Three Fiber pitfalls. - Do not create objects every frame. new Vector3() inside useFrame produces garbage, and the garbage collector pauses the page to clean it up. Create helpers once, outside the loop, and reuse them. - Keep physics small. Use simple collider shapes, and let objects that do not move be fixed bodies.","u":"/blog/webxr-performance-budget-quest-3#reduce-javascript-work-per-frame"},{"k":"blog","t":"A performance budget for WebXR on Quest 3","h":"Textures and loading","x":"Large textures use memory and take time to upload, which shows up as a hitch when an object first appears. Size textures for how big they are seen, not for how they were exported. GPU-compressed textures in the KTX 2.0 format stay compressed in graphics memory. Load what the first view needs before the session starts.","u":"/blog/webxr-performance-budget-quest-3#textures-and-loading"},{"k":"blog","t":"A performance budget for WebXR on Quest 3","h":"A routine that works","x":"1. Decide the frame rate you will hold, and so your time per frame. 2. Test on the headset early, with the real content, not at the end. 3. When frames drop, run the small-picture experiment to find the side that is over budget. 4. Fix the biggest cost on that side. Measure again. 5. Keep the draw call and triangle counts you ended with as the budget for the scene.","u":"/blog/webxr-performance-budget-quest-3#a-routine-that-works"},{"k":"blog","t":"A performance budget for WebXR on Quest 3","h":"Doing this in Graspable","x":"Graspable checks that a scene builds, loads without errors, and still passes the headset sessions you recorded. It does not measure frame rate on your headset. An emulated headset on a desktop computer cannot tell you that, and we would rather say so than show a number that means nothing. What it does make quick is the loop around the measurement. Open the project on your headset from the preview ","s":"Graspable checks that a scene builds, loads without errors, and still passes the headset sessions you recorded. It does not measure frame rate on your headset. An emulated headset on a desktop computer cannot tell you that, and we would rather say so than show a number that means nothing. What it does make quick is the loop around the measurement. Open the project on your headset from the preview and edits keep hot-reloading there. Ask the agent for a specific change, for example \"draw the trees as one instanced mesh\" or \"print draw calls and triangles every two seconds\", and measure again on the device.","u":"/blog/webxr-performance-budget-quest-3#doing-this-in-graspable"},{"k":"blog","t":"A performance budget for WebXR on Quest 3","h":"Sources","x":"- WebXR performance optimization workflow, WebXR app framerate control and WebXR fixed foveated rendering, Meta - WebXR performance guide and XRWebGLLayer.fixedFoveation, MDN - WebXR Device API, W3C - Scaling performance and performance pitfalls, React Three Fiber - Performance, @react-three/xr - WebGLRenderer.info and InstancedMesh, three.js","u":"/blog/webxr-performance-budget-quest-3#sources"},{"k":"blog","t":"Describe a scene, step inside it","x":"Graspable is now a desktop studio. You say what you want, an agent builds it as a real project on your computer, and you try it on a headset. Making something for a headset has always started with a long list of chores. Pick a framework. Set up a build. Learn how controllers, hands and room scanning work. Find out on the device that it does not work. Graspable starts somewhere else. You describe t","s":"Graspable is now a desktop studio. You say what you want, an agent builds it as a real project on your computer, and you try it on a headset. Making something for a headset has always started with a long list of chores. Pick a framework. Set up a build. Learn how controllers, hands and room scanning work. Find out on the device that it does not work. Graspable starts somewhere else. You describe the scene. Request: Build an AR measuring tape. I point at a surface, pull the trigger to place points, and see the length in centimetres. Agent: I'll build it in App.jsx with a reticle that follows the surface, then check it in the emulated headset.","u":"/blog/describe-a-scene-step-inside-it"},{"k":"blog","t":"Describe a scene, step inside it","h":"A real project, on your computer","x":"The agent does not hand you a preview you cannot keep. It edits a normal project folder on your disk: React, three.js and WebXR, the same tools professionals use. You can open it in any editor, and it is yours. While it works you see everything: each file it changes, each command it runs, and what it cost. Anything risky asks you first. If you do not like the result, undo the whole run with one cl","s":"The agent does not hand you a preview you cannot keep. It edits a normal project folder on your disk: React, three.js and WebXR, the same tools professionals use. You can open it in any editor, and it is yours. While it works you see everything: each file it changes, each command it runs, and what it cost. Anything risky asks you first. If you do not like the result, undo the whole run with one click. Code, live preview, and everything the agent did.","u":"/blog/describe-a-scene-step-inside-it#a-real-project-on-your-computer"},{"k":"blog","t":"Describe a scene, step inside it","h":"It has to run before it says done","x":"An agent that says \"done\" when nothing works is worse than no agent. So a run in Graspable only finishes when the project builds and loads in a real browser without errors. When it fails, the error goes back to the agent, up to three times.","u":"/blog/describe-a-scene-step-inside-it#it-has-to-run-before-it-says-done"},{"k":"blog","t":"Describe a scene, step inside it","h":"A headset you do not have to put on","x":"The app includes an emulated headset standing in a scanned room. You can look around, point controllers, and try the scene before you reach for the real device. This is the measuring tape from the request above, as the agent built it: The emulated Quest 3 in a scanned living room. When you are ready, open the same scene on your Quest or Vision Pro over Wi-Fi or USB.","u":"/blog/describe-a-scene-step-inside-it#a-headset-you-do-not-have-to-put-on"},{"k":"blog","t":"Describe a scene, step inside it","h":"Your AI account","x":"Graspable does not sell you a model. Sign in with your ChatGPT subscription, use Claude Code with your Claude subscription, or add an API key. Building on your own computer is free. Download Graspable and describe your first scene.","u":"/blog/describe-a-scene-step-inside-it#your-ai-account"},{"k":"blog","t":"How Graspable checks its own work","x":"A run is not finished until the scene builds, loads in a browser and still passes the headset sessions you recorded. Anyone who has worked with a coding agent knows the moment. It says the work is finished. You open the result and the screen is blank. We decided the agent does not get to decide when it is done. Three checks do.","u":"/blog/how-graspable-checks-its-own-work"},{"k":"blog","t":"How Graspable checks its own work","h":"1. It builds","x":"After the agent stops, Graspable builds the project. A build error goes straight back to the agent with the message, and it tries again.","u":"/blog/how-graspable-checks-its-own-work#1-it-builds"},{"k":"blog","t":"How Graspable checks its own work","h":"2. It loads","x":"A scene can build and still crash the moment it starts. So the built scene is opened in a real browser on your computer, with no window, and watched for errors. You get a screenshot of what it rendered.","u":"/blog/how-graspable-checks-its-own-work#2-it-loads"},{"k":"blog","t":"How Graspable checks its own work","h":"3. It still does what you recorded","x":"This is the part that is special to headsets. In the emulated headset you can record a session: where you looked, what you pointed at, which buttons you pressed. Graspable saves it as a test. After every later change, that session is replayed. If pulling the trigger used to place a point and now does nothing, you hear about it before you put the headset on. A game built in one request, checked in ","s":"This is the part that is special to headsets. In the emulated headset you can record a session: where you looked, what you pointed at, which buttons you pressed. Graspable saves it as a test. After every later change, that session is replayed. If pulling the trigger used to place a point and now does nothing, you hear about it before you put the headset on. A game built in one request, checked in the emulated headset. The agent can record sessions too, to prove an interaction works. It can never edit a session you recorded, and a recording that captured nothing is not kept.","u":"/blog/how-graspable-checks-its-own-work#3-it-still-does-what-you-recorded"},{"k":"blog","t":"How Graspable checks its own work","h":"Three tries, then you","x":"When a check fails, the agent gets the error and up to three attempts to repair it. If it still fails, the run stops and tells you why. It does not loop forever on your account. When the scene looks different from your recording, that is not treated as a failure. You are shown the difference and you decide: accept the new look, or ask for a fix. Read how to record your first test.","u":"/blog/how-graspable-checks-its-own-work#three-tries-then-you"},{"k":"made","t":"The National Theatre in Prague, walked through together","x":"Lukáš Eršil’s open 3D model of the Národní divadlo, turned into a place for several people at once: walk the street, the foyer and the auditorium in a headset or a browser, dress for 1883, talk, and wind the building back to its foundations","u":"/a/kkqf82wc"},{"k":"made","t":"A penthouse sales suite in Unreal Engine","x":"The living room of an apartment that is not built yet, for buyers to walk through in a headset: low sun across an oak floor, a wall of glass onto a terrace, and a lever that changes day to evening.","u":"/a/tbmqne63"},{"k":"made","t":"A tape measure in a real room","x":"Point at any surface, pull the trigger, and read the distance on a steel tape.","u":"/a/ifz7klc6"},{"k":"made","t":"Arcade basketball","x":"Grab a ball off the rack, swing, let go: it bounces off the backboard, rattles the rim and the scoreboard counts.","u":"/a/s8t7x4oc"},{"k":"made","t":"A solar system you can pick up","x":"Planets circle the Sun on brass rings in a planetarium; grab one and its lesson card opens beside it.","u":"/a/o2o4kb1i"},{"k":"made","t":"Chair configurator","x":"Point at a swatch and pull the trigger: shell colour, seat pad and legs change on the chair, the price adds up, and a summary lists the choice.","u":"/a/0bawi3nu"},{"k":"made","t":"Greenhouse tour","x":"A five-stop walk through a generated Victorian greenhouse at dusk: Next takes you to the next stop and its plate.","u":"/a/ww40msdw"},{"k":"made","t":"Loft viewing","x":"A viewing of the loft with six info points and five stops, ending at the sofa with the asking price and a Book a visit button.","u":"/a/i88ocd4i"},{"k":"made","t":"Tavern tankards, three rounds","x":"Throw balls at pewter tankards on the bar of a generated tavern: three rounds against the clock, swinging targets, a bucket worth five.","u":"/a/5d5botjh"},{"k":"made","t":"Cannonballs on a pirate deck","x":"On the deck of a generated pirate ship, throw cannonballs at green bottles on the barrels: ten points each, three rounds, a bell at the end.","u":"/a/1ocwf0q9"},{"k":"made","t":"Chime bars for bare hands","x":"Eight coloured bars on a walnut frame: touch one with a fingertip and it dips, glows and rings its note.","u":"/a/78anufvg"},{"k":"made","t":"Ceramics gallery","x":"Seven glazed pieces on the worktable of a generated loft: point at one and it lifts, turns and shows its card.","u":"/a/wlw07saj"},{"k":"made","t":"Furnish a scanned room","x":"Pick a piece from the palette, point at the floor of a real scanned room to place it, and see its distance to the wall and to its neighbours.","u":"/a/6wquwdau"},{"k":"made","t":"A work-order console for technicians","x":"Pick a work order, answer its inspection checklist with Pass, Fail or N/A, watch the asset's readings, and sign off: a panel that floats beside the machine.","u":"/a/88f3zr6u"},{"k":"made","t":"A shelving configurator for Apple Vision Pro","x":"Choose width, height, finish and extras in a window, read the parts and the price, and see the shelf at real size in your room.","u":"/a/pae5d4ah"},{"k":"made","t":"A guided shutdown of a pump unit","x":"A five-step maintenance procedure in mixed reality: the step card says what to do, the part to act on glows, and the unit answers.","u":"/a/c8y90muw"},{"k":"made","t":"A design review of a gearbox","x":"A gearbox at real size that comes apart: explode it, cut it open, select a part to read its material and mass, and pin notes to it.","u":"/a/dzv4o9cq"},{"k":"made","t":"Pick by vision for a warehouse","x":"The next pick stays in view on a strip that follows your head, the right bin lights up, and a pinch confirms it.","u":"/a/opduwujd"},{"k":"made","t":"A furniture catalogue from the internet, in your room","x":"The catalogue is loaded live: pick a scanned piece from the list, see it at its true size, and place several in your own room.","u":"/a/oro7hk1z"},{"k":"made","t":"A kitchen island showroom in Unreal Engine","x":"A kitchen island at real size with an oak worktop, a sink, bar stools and pendant lamps, written as a script and built by Unreal Engine.","u":"/a/k4am2tqi"},{"k":"made","t":"A machine start-up station in Unreal Engine","x":"A plant-room console for training operators: pull the main power lever down and the work lamp comes on and the sign reads POWER ON.","u":"/a/olyzk6ka"},{"k":"template","t":"React XR","x":"A small working scene for VR and AR: things to pick up and throw, buttons, readable text, physics and sound. The starting point for anything.","u":"/templates/react-xr"},{"k":"template","t":"Splat World","x":"Stand inside a photorealistic generated place at real scale, look around, teleport, and add your own things to it.","u":"/templates/splat-world"},{"k":"template","t":"World Explorer","x":"Walk through a generated place as if it were solid: floors hold you, walls and furniture stop you, with info points and a guided tour.","u":"/templates/world-explorer"},{"k":"template","t":"World Game","x":"A game inside a generated place: throw balls that bounce off the real room, knock mugs off the bar, ring the targets, keep score.","u":"/templates/world-game"},{"k":"template","t":"WebGPU Splats","x":"Millions of Gaussian splats on WebGPU: load captured places and make splats with code that runs on the graphics card, in a browser or a headset.","u":"/templates/webgpu-splats"},{"k":"template","t":"Scanned Room","x":"Start inside a 3D capture of a real room, where what you add sits on the real floor, walls and furniture at real scale.","u":"/templates/scanned-room"},{"k":"template","t":"Product Studio","x":"Show one product in a photo studio, let people choose its colours and materials, and place it at real size in VR or in their own room.","u":"/templates/product-studio"},{"k":"template","t":"XR Game","x":"A physics play space for VR games: grab, throw, hit and score, with a scoreboard, sounds and a can-knockdown game to start from.","u":"/templates/xr-game"},{"k":"template","t":"Hands & Spatial UI","x":"Panels, buttons and sliders you pinch and poke with your bare hands, ready to become a dashboard, an instrument or a tool.","u":"/templates/spatial-ui"},{"k":"template","t":"Cross-platform XR","x":"One app for every headset, pair of glasses, phone and computer: a product shown in your own room where the device has passthrough, in a photographed room where it does not.","u":"/templates/crossplatform"},{"k":"template","t":"Display glasses web app","x":"An app that runs on Meta Ray-Ban Display glasses as a web page: 600 by 600 pixels, moved through with the Neural Band or the touchpad, built with Meta's UI Toolkit. Publish it and the glasses install it from a code. Each screen also offers ","u":"/templates/display-glasses"},{"k":"template","t":"Expo panel app","x":"A flat app that opens as a floating panel on Meta Quest, Meta VR Glasses and Android XR, and as a web page everywhere else: big targets, a layout that fits a window, built with Expo.","u":"/templates/expo-quest"},{"k":"template","t":"Glasses field notes","x":"Look through your glasses on your phone, save a moment with a tap on the temple, say a note and it is written under the photo, and hear the answer in the glasses' own speakers. A finished-looking app for Ray-Ban Meta Gen 1 and Gen 2 and Oak","u":"/templates/glasses-journal"},{"k":"template","t":"Meta glasses app","x":"A phone app that talks to Meta AI glasses through the Meta Wearables Device Access Toolkit: the camera stream and photos, a screen on Meta Ray-Ban Display, inputs, motion, speech and voice launch, with mock glasses for development. Everythi","u":"/templates/meta-glasses"},{"k":"template","t":"Multi-user","x":"A room several people open at once: everyone sees each other, picks up the same objects and presses the same board.","u":"/templates/multiuser"},{"k":"template","t":"Unity for Android XR","x":"A native Unity project for Android XR headsets and glasses (Samsung Galaxy XR, XREAL Project Aura and others): passthrough, real surfaces, hands and gaze, built as an Android app you install on the device.","u":"/templates/unity-androidxr"},{"k":"template","t":"Unity AR tabletop","x":"Place a miniature rescue mission on your table. Restore three power relays, guide a small robot and watch the courier return. Includes an editable short film.","u":"/templates/unity-ar"},{"k":"template","t":"Unity for Quest","x":"A native Unity project for Meta Quest and Meta VR Glasses: mixed reality out of the box, with passthrough, real surfaces, hands, gaze and controllers, built as an Android app you install on the headset.","u":"/templates/unity-quest"},{"k":"template","t":"Unity for XREAL","x":"A native Unity project for XREAL glasses: Project Aura through Android XR out of the box, XREAL One and Air through the XREAL SDK you add; hands, gaze and the real room, built as an Android app.","u":"/templates/unity-xreal"},{"k":"template","t":"Unreal Engine XR","x":"An Unreal Engine 5 project for OpenXR headsets whose level is a Python script: scanned furniture and props at real size, daylight that bounces, things to grab and switch, Blueprints the agent writes.","u":"/templates/unreal-xr"},{"k":"template","t":"Apple Vision Pro","x":"A native Apple Vision Pro app: a window with controls and an immersive space around the person, built and run in the visionOS Simulator by your own Xcode.","u":"/templates/visionos"},{"k":"template","t":"Tabletop AR","x":"A miniature sci-fi rescue: watch the film, then turn the relays to guide a ship home. Place it on a surface in WebXR or on an image target in a phone browser.","u":"/templates/web-ar"},{"k":"skill","t":"3D Spatial Animation","x":"Use when working in Blender, Unity 3D, Unreal Engine, Cinema 4D, VR/AR applications, or any three-dimensional animation work. dylantarre Blender Unity Unreal Engine animation xr","u":"/skills/3d-spatial"},{"k":"skill","t":"3D Testing and Visual QA","x":"Test Three.js features with browser checks, deterministic visual fixtures, lifecycle tests, and device-specific evidence. cesartevisual Three.js 3d performance","u":"/skills/threejs-testing"},{"k":"skill","t":"Accessible 3D Experiences","x":"Make Three.js interactions usable with keyboards, assistive technology, reduced motion, and non-3D fallbacks. cesartevisual Three.js 3d accessibility","u":"/skills/threejs-accessibility"},{"k":"skill","t":"Accessible Motion Design","x":"Use when implementing reduced motion alternatives, vestibular-safe animations, WCAG compliance, or designing for users with motion sensitivity. dylantarre  animation accessibility","u":"/skills/accessible-motion"},{"k":"skill","t":"Animation and Timelines","x":"Implement Three.js animation clips, skeletal animation, transitions, scroll timelines, and frame-independent motion. cesartevisual Three.js 3d animation","u":"/skills/threejs-animation"},{"k":"skill","t":"Articulated Gaussian-splat scenes","x":"Reason about parts, joints and interaction in captured splat scenes. Research guidance with upstream tooling requirements. jaccen  splats","u":"/skills/3dgs-articulated-reasoner"},{"k":"skill","t":"Asset Loading and Optimization","x":"Load, validate, compress, cache, and release glTF models, textures, decoder assets, and user-supplied 3D files. cesartevisual Three.js 3d","u":"/skills/threejs-assets"},{"k":"skill","t":"Beat-synced editing","x":"Plan cuts, transitions, speed ramps and shot timing around the rhythm of a piece. iart-ai  film animation","u":"/skills/beat-sync-editing"},{"k":"skill","t":"BIM, GIS, and Large Coordinates","x":"Integrate Three.js BIM/IFC, GIS, CAD, measurements, clipping, and large-coordinate datasets. cesartevisual Three.js 3d xr","u":"/skills/threejs-spatial-data"},{"k":"skill","t":"Blender animation","x":"Create keyframes, actions, NLA clips and retargeted motion, then bake and export animation for a game engine. EL4CTEO Blender assets animation","u":"/skills/blender-animation"},{"k":"skill","t":"Blender asset checks","x":"Check mesh topology, normals, UVs, scale and intersections before handing an asset to the next production stage. EL4CTEO Blender assets performance","u":"/skills/blender-asset-qa"},{"k":"skill","t":"Blender assets for real-time scenes","x":"Set geometry, material, texture and bone budgets; build levels of detail and reduce draw calls. EL4CTEO Blender assets performance","u":"/skills/blender-game-optimization"},{"k":"skill","t":"Blender character rigging","x":"Fit a Rigify rig, set up IK and FK, skin a character and validate deformation before engine export. EL4CTEO Blender assets animation","u":"/skills/blender-character-rigging"},{"k":"skill","t":"Blender export pipelines","x":"Prepare and check FBX, GLB or USD exports with the right scale, axes, skeleton and animation clips. EL4CTEO Blender assets","u":"/skills/blender-export-pipelines"},{"k":"skill","t":"Blender facial rigs and shape keys","x":"Build expressions, speech shapes and corrective deformations with blendshapes, bones and drivers. EL4CTEO Blender assets animation","u":"/skills/blender-facial-shapekeys"},{"k":"skill","t":"Blender foundations","x":"Drive Blender through Python or a connected editor, check the installed API and verify the result. EL4CTEO Blender assets","u":"/skills/blender-core"},{"k":"skill","t":"Blender lighting and look development","x":"Light characters and products, compose cameras and tune exposure and materials before rendering. EL4CTEO Blender assets","u":"/skills/blender-lighting-lookdev"},{"k":"skill","t":"Blender PBR materials","x":"Build physically based materials with correct texture color spaces, channel packing, normals and surface values. EL4CTEO Blender assets","u":"/skills/blender-pbr-materials"},{"k":"skill","t":"Blender rendering","x":"Choose render settings, lighting passes, denoising and color management for stills and animation. EL4CTEO Blender assets","u":"/skills/blender-rendering"},{"k":"skill","t":"Blender retopology and UVs","x":"Prepare clean deformation topology, unwrap UVs and set consistent texture density for generated or scanned assets. EL4CTEO Blender assets","u":"/skills/blender-retopology-uv"},{"k":"skill","t":"Blender texture baking","x":"Bake normals and surface maps from detailed assets into textures suitable for real-time rendering. EL4CTEO Blender assets","u":"/skills/blender-texture-baking"},{"k":"skill","t":"Build and Delivery","x":"Prepare and verify Three.js web builds, asset hosting, decoder paths, caching, and deployment compatibility. cesartevisual Three.js 3d","u":"/skills/threejs-deployment"},{"k":"skill","t":"Camera capture on AI glasses","x":"Build a deliberate capture flow with permission, progress and a useful saved result. Graspable  glasses","u":"/skills/glasses-camera-capture"},{"k":"skill","t":"Color in motion","x":"Choose a restrained motion palette and avoid muddy color transitions when animating or grading a film. iart-ai  film animation","u":"/skills/color-motion"},{"k":"skill","t":"Custom Shaders","x":"Write and debug Three.js GLSL shaders, custom material effects, uniforms, and shader deformation. cesartevisual Three.js 3d","u":"/skills/threejs-shaders"},{"k":"skill","t":"Data Visualization Animation","x":"Use when animating charts, graphs, dashboards, data transitions, or any information visualization work. dylantarre  animation","u":"/skills/data-visualization"},{"k":"skill","t":"Dialogue systems","x":"Design branching conversations with conditions, variables, localization and a suitable authoring format. gamedev-skills  games","u":"/skills/dialogue-systems"},{"k":"skill","t":"Game camera systems","x":"Build follow, orbit and first-person cameras with collision, framing and controlled smoothing. gamedev-skills  games","u":"/skills/camera-systems"},{"k":"skill","t":"Game Development Animation","x":"Use when implementing game animations, player feedback, character movement, or interactive entertainment in Unity, Unreal, or other game engines. dylantarre Blender Unity Unreal Engine animation","u":"/skills/game-development"},{"k":"skill","t":"Game performance optimization","x":"Profile frame time, identify the CPU or GPU bottleneck and choose measurable fixes. gamedev-skills  games performance","u":"/skills/performance-optimization"},{"k":"skill","t":"Gaussian-splat compression and deployment","x":"Compare quantization, pruning, streaming and levels of detail for splat scenes on the web and mobile devices. jaccen  splats performance","u":"/skills/3dgs-compression-deploy"},{"k":"skill","t":"Gaussian-splat engineering guide","x":"Choose a practical path from Gaussian-splatting research to a deployed application. jaccen  splats","u":"/skills/3dgs-engineering-guide"},{"k":"skill","t":"Geometry and Instancing","x":"Create or optimize Three.js BufferGeometry, procedural meshes, instanced objects, and geometry attributes. cesartevisual Three.js 3d","u":"/skills/threejs-geometry"},{"k":"skill","t":"Grab, hold and throw in VR","x":"Implement controller grabbing, object attachment, release velocity and feedback in a Three.js scene. pavlin-dev Three.js WebXR xr","u":"/skills/vr-grab-interaction"},{"k":"skill","t":"GSAP foundations","x":"Animate with tweens, easing, staggered timing and responsive or reduced-motion variants. greensock GSAP animation","u":"/skills/gsap-core"},{"k":"skill","t":"GSAP performance","x":"Reduce animation jank with transform-based movement, batching and fewer forced layout calculations. greensock GSAP animation performance","u":"/skills/gsap-performance"},{"k":"skill","t":"GSAP timelines","x":"Sequence animations with positions, labels, nested timelines and playback controls. greensock GSAP animation","u":"/skills/gsap-timeline"},{"k":"skill","t":"GSAP with React","x":"Use GSAP in React with scoped animations, refs and cleanup that follows the component lifecycle. greensock GSAP animation","u":"/skills/gsap-react"},{"k":"skill","t":"HyperFrames motion reel","x":"Plan and render beat-synced motion graphics with HTML, GSAP, SVG and Canvas. Instructions are in Japanese. Sunwood-ai-labs HyperFrames film","u":"/skills/hyperframes-motion-reel"},{"k":"skill","t":"Immersive film review","x":"Check story, camera comfort, audio and viewing controls before sharing a spatial film. Graspable  film xr performance","u":"/skills/immersive-film-review"},{"k":"skill","t":"Lighting and Shadows","x":"Light Three.js scenes using environment maps, direct lights, baked light, and tuned shadow maps. cesartevisual Three.js 3d","u":"/skills/threejs-lighting"},{"k":"skill","t":"Materials and Textures","x":"Implement Three.js PBR materials, texture mapping, transparency, and color-management fixes. cesartevisual Three.js 3d","u":"/skills/threejs-materials"},{"k":"skill","t":"Micro-interaction Animation","x":"Use when designing small UI feedback moments like button states, toggles, form validation, loading indicators, or notification badges. dylantarre  animation","u":"/skills/micro-interactions"},{"k":"skill","t":"Motion design principles","x":"Plan animation timing, easing and choreography before implementing the motion. LottieFiles  animation","u":"/skills/lottie-motion-design"},{"k":"skill","t":"Particles and Large Data","x":"Build Three.js particle systems, point clouds, large repeated datasets, and CPU or GPU simulations. cesartevisual Three.js 3d","u":"/skills/threejs-particles"},{"k":"skill","t":"Performance and Memory","x":"Profile and optimize Three.js frame time, GPU workload, asset memory, and resource leaks. cesartevisual Three.js 3d performance","u":"/skills/threejs-performance"},{"k":"skill","t":"Physics Integration","x":"Integrate Rapier or another physics engine with Three.js, including colliders, fixed timesteps, and render synchronization. cesartevisual Three.js 3d","u":"/skills/threejs-physics"},{"k":"skill","t":"Picking and Controls","x":"Implement Three.js pointer picking, orbit controls, object dragging, selection, and synchronized HTML controls. cesartevisual Three.js 3d","u":"/skills/threejs-interaction"},{"k":"skill","t":"Postprocessing Pipelines","x":"Add or debug Three.js bloom, antialiasing, ambient occlusion, depth effects, and render-target composition. cesartevisual Three.js 3d","u":"/skills/threejs-postprocessing"},{"k":"skill","t":"Product Viewers and Configurators","x":"Build Three.js product viewers with material variants, camera presets, hotspots, configuration state, and image export. cesartevisual Three.js 3d","u":"/skills/threejs-product-viewer"},{"k":"skill","t":"React Three Fiber and Drei","x":"Build or fix React Three Fiber scenes, Drei helpers, Suspense loading, events, and React lifecycle integration. cesartevisual Three.js 3d","u":"/skills/threejs-r3f"},{"k":"skill","t":"Rendering Diagnostics","x":"Diagnose Three.js black canvases, missing meshes, wrong colors, flicker, picking bugs, and context loss. cesartevisual Three.js 3d performance","u":"/skills/threejs-debugging"},{"k":"skill","t":"Scenes, Cameras, and Math","x":"Build or correct Three.js scene hierarchies, camera framing, coordinate transforms, and responsive renderer lifecycles. cesartevisual Three.js 3d","u":"/skills/threejs-scene"},{"k":"skill","t":"Shot composition","x":"Frame action, guide attention and plan camera moves, with safe areas for different video formats. iart-ai  film animation","u":"/skills/shot-composition"},{"k":"skill","t":"Small-display glasses interfaces","x":"Design readable screens and voice actions for glasses with a compact display. Graspable  glasses accessibility","u":"/skills/glasses-small-display"},{"k":"skill","t":"Spatial Audio","x":"Add Three.js spatial audio, user-controlled playback, listener movement, and audio lifecycle handling. cesartevisual Three.js 3d","u":"/skills/threejs-audio"},{"k":"skill","t":"Tabletop AR placement","x":"Build placement and recovery flows for phones and passthrough headsets. Graspable  xr games","u":"/skills/tabletop-ar-placement"},{"k":"skill","t":"Three.js Web Projects","x":"Plan and implement complete Three.js web experiences, or select focused skills for an existing 3D feature. cesartevisual Three.js 3d","u":"/skills/threejs-web"},{"k":"skill","t":"Unity animation","x":"Build Animator states, transitions, blend trees, layers and humanoid IK for character movement. gamedev-skills Unity games animation","u":"/skills/unity-animation"},{"k":"skill","t":"Unity input","x":"Connect actions, action maps and player input, with callbacks or polling suited to the project. gamedev-skills Unity games","u":"/skills/unity-input-system"},{"k":"skill","t":"Unity physics","x":"Set up rigid bodies, collision layers, raycasts, triggers and joints for a 3D scene. gamedev-skills Unity games","u":"/skills/unity-physics"},{"k":"skill","t":"Unreal Blueprints","x":"Structure gameplay with Blueprint classes, events, functions and communication between actors. gamedev-skills Unreal Engine games","u":"/skills/unreal-blueprints"},{"k":"skill","t":"Unreal Enhanced Input","x":"Connect input actions, mapping contexts, triggers and modifiers to gameplay. gamedev-skills Unreal Engine games","u":"/skills/unreal-enhanced-input"},{"k":"skill","t":"Unreal Niagara effects","x":"Build Niagara systems and emitters, expose parameters and control effects from gameplay. gamedev-skills Unreal Engine games","u":"/skills/unreal-niagara"},{"k":"skill","t":"Video Motion Graphics","x":"Use when creating After Effects compositions, Premiere Pro motion, video titles, explainer videos, or broadcast motion graphics. dylantarre  animation","u":"/skills/video-motion-graphics"},{"k":"skill","t":"Voice interactions for AI glasses","x":"Design short, interruptible voice interactions with visible and spoken feedback. Graspable  glasses accessibility","u":"/skills/glasses-voice-interactions"},{"k":"skill","t":"VR comfort and polish","x":"Review feedback, timing, haptics and motion comfort before calling a VR experience finished. pavlin-dev Three.js WebXR xr","u":"/skills/vr-polish-checklist"},{"k":"skill","t":"VR environment design","x":"Build a readable, grounded WebXR environment with lighting, materials, atmosphere and a sensible scale. pavlin-dev Three.js WebXR xr","u":"/skills/environment-scaffold"},{"k":"skill","t":"VR particles and effects","x":"Create pooled, instanced effects that read correctly in stereo and fit a headset frame budget. pavlin-dev Three.js WebXR xr","u":"/skills/vfx-and-particles"},{"k":"skill","t":"Web and Framework Integration","x":"Integrate Three.js with responsive HTML, Next.js or other frameworks, scroll sections, overlays, and multiple views. cesartevisual Three.js 3d","u":"/skills/threejs-html"},{"k":"skill","t":"Web Motion Design","x":"Use when building CSS animations, JavaScript transitions, React/Vue motion, or any browser-based animation work. dylantarre  animation","u":"/skills/web-motion-design"},{"k":"skill","t":"WebGPU and TSL","x":"Build or migrate Three.js WebGPURenderer, TSL node materials, and GPU compute features. cesartevisual Three.js 3d","u":"/skills/threejs-webgpu"},{"k":"skill","t":"WebXR","x":"Implement Three.js VR or AR sessions, XR input, reference spaces, hit testing, and session lifecycle. cesartevisual Three.js 3d xr","u":"/skills/threejs-xr"},{"k":"skill","t":"WebXR session setup","x":"Set up immersive sessions and the XR render loop, with recovery and a useful desktop fallback. pavlin-dev Three.js WebXR xr","u":"/skills/xr-session-setup"},{"k":"use","t":"Product configurators and showrooms","x":"Let buyers see the product at real size, in their own room, in the finish they choose. Build a product configurator people try at real size in their own room: materials, sizes and options, in the browser, on a headset or on Apple Vision Pro","u":"/use/product-configurators"},{"k":"use","t":"Real estate and architecture","x":"Walk buyers and clients through a space before it is built, or furnish one that already is. Build property viewings, sales suites and room planners people walk through in a headset or a browser, from your models, scans or a described space.","u":"/use/real-estate-and-architecture"},{"k":"use","t":"Industrial training and operations","x":"Teach a procedure at the machine, or before anyone has to touch it. Build step-by-step XR training, maintenance procedures, design reviews and warehouse guidance for headsets and AR glasses. Made by describing it, as a project your team own","u":"/use/industrial-training"},{"k":"use","t":"Culture, museums and tourism","x":"Open a building, a collection or a place to people who cannot be there, together. Build virtual galleries, guided tours and shared walkthroughs of real places, for visitors in a headset or a browser. Made by describing it, as a project your","u":"/use/culture-and-museums"},{"k":"use","t":"Education","x":"Make the thing on the slide something a student can pick up and turn. Build lessons students can step into: models they pick up, experiments they run, instruments they play. In a headset or a browser, made by describing it, as a project you","u":"/use/education"},{"k":"use","t":"Brand experiences, games and events","x":"Give people at a stand or a launch something to do with their hands. Build short XR games and brand experiences for launches, stands and campaigns, that open from a link in a headset or a browser. Made by describing it, as a project your st","u":"/use/brand-experiences-and-events"},{"k":"comparison","t":"Graspable vs Replit","h":"Compare","x":"Both accept a description of an XR experience. Do you prefer a local project with integrated XR checks, or a cloud workspace with hosting? Replit is an AI app builder with a cloud workspace. Its VR guide explicitly describes creating WebXR projects with Replit Agent.","u":"/compare/replit"},{"k":"comparison","t":"Graspable vs Replit","h":"What you build · Graspable","x":"A desktop app for Windows, macOS and Linux. Describe a 3D or XR scene, typed or spoken; an agent edits a real project on your computer: React Three Fiber and WebXR, or a native Unity, Unreal Engine or visionOS project.","u":"/compare/replit#fact-what-you-build"},{"k":"comparison","t":"Graspable vs Replit","h":"What you build · Replit","x":"WebXR apps using Three.js, Babylon.js, A-Frame or React Three Fiber are documented in the VR builder guide.","u":"/compare/replit#fact-what-you-build"},{"k":"comparison","t":"Graspable vs Replit","h":"The AI · Graspable","x":"Your own AI account: an API key from Anthropic, OpenAI, OpenRouter and others, a ChatGPT subscription, or Claude Code with a Claude subscription. Models come from the provider’s own list.","u":"/compare/replit#fact-the-ai"},{"k":"comparison","t":"Graspable vs Replit","h":"The AI · Replit","x":"Replit Agent routes between models automatically. Core and Pro users can choose available models in Power or Max mode; Free mode uses automatic routing.","u":"/compare/replit#fact-the-ai"},{"k":"comparison","t":"Graspable vs Replit","h":"Where it runs · Graspable","x":"On your computer. Projects, previews, builds, keys and sign-ins stay there. A paid plan can move a project to a Cloud Agent and back.","u":"/compare/replit#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs Replit","h":"Where it runs · Replit","x":"A cloud workspace with browser preview and deployment.","u":"/compare/replit#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs Replit","h":"Checking the result · Graspable","x":"Web projects must build and load in a browser before a run completes. Native templates use their engine’s checks. Recorded headset tests are replayed when configured.","u":"/compare/replit#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs Replit","h":"Checking the result · Replit","x":"App Testing exercises supported web apps in a browser and attempts repairs. The VR guide describes opening the deployed HTTPS URL in a compatible headset browser. Those docs do not establish whether headset emulation is available elsewhere in the product.","u":"/compare/replit#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs Replit","h":"Your code · Graspable","x":"Editable project files on your disk: React Three Fiber and WebXR for web projects, or the selected native engine’s project. File history and undo are available.","u":"/compare/replit#fact-your-code"},{"k":"comparison","t":"Graspable vs Replit","h":"Your code · Replit","x":"The Git pane supports commits, branches, merges and connections to remote repositories, including GitHub.","u":"/compare/replit#fact-your-code"},{"k":"comparison","t":"Graspable vs Replit","h":"Price · Graspable","x":"Local building with your own supported AI account is free. Monthly plans: Builder $20, Pro $40, Studio $99 per seat with a three-seat minimum. Features and Credits differ by plan (graspable.dev/pricing).","u":"/compare/replit#fact-price"},{"k":"comparison","t":"Graspable vs Replit","h":"Price · Replit","x":"See the current Replit pricing page and usage terms. Compare the plan and usage charges for your intended workload.","u":"/compare/replit#fact-price"},{"k":"comparison","t":"Graspable vs v0 by Vercel","h":"Compare","x":"With Meta’s template, v0 can take a prompt to a WebXR scene you open on a Quest. Graspable does the same on your computer with an emulated headset and a build gate. Which one, and when? v0 is Vercel’s browser app builder: it generates Next.js and React apps, runs them in a Vercel sandbox and deploys them to Vercel. Meta published an official Immersive Web SDK template for it at Connect 2026.","u":"/compare/v0"},{"k":"comparison","t":"Graspable vs v0 by Vercel","h":"What you build · Graspable","x":"A desktop app for Windows, macOS and Linux. Describe a 3D or XR scene, typed or spoken; an agent edits a real project on your computer: React Three Fiber and WebXR, or a native Unity, Unreal Engine or visionOS project.","u":"/compare/v0#fact-what-you-build"},{"k":"comparison","t":"Graspable vs v0 by Vercel","h":"What you build · v0 by Vercel","x":"Next.js and React web apps. WebXR through the \"Immersive Web SDK Starter App\" template from Meta (24 September 2026), a static Vite app with the IWER emulator in development.","u":"/compare/v0#fact-what-you-build"},{"k":"comparison","t":"Graspable vs v0 by Vercel","h":"The AI · Graspable","x":"Your own AI account: an API key from Anthropic, OpenAI, OpenRouter and others, a ChatGPT subscription, or Claude Code with a Claude subscription. Models come from the provider’s own list.","u":"/compare/v0#fact-the-ai"},{"k":"comparison","t":"Graspable vs v0 by Vercel","h":"The AI · v0 by Vercel","x":"v0’s own composite models (Mini, Pro, Max, Max Fast); the frontier models behind them are not named. No bring-your-own key found.","u":"/compare/v0#fact-the-ai"},{"k":"comparison","t":"Graspable vs v0 by Vercel","h":"Where it runs · Graspable","x":"On your computer. Projects, previews, builds, keys and sign-ins stay there. A paid plan can move a project to a Cloud Agent and back.","u":"/compare/v0#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs v0 by Vercel","h":"Where it runs · v0 by Vercel","x":"In the browser and a Vercel sandbox; deploys to Vercel.","u":"/compare/v0#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs v0 by Vercel","h":"Checking the result · Graspable","x":"Web projects must build and load in a browser before a run completes. Native templates use their engine’s checks. Recorded headset tests are replayed when configured.","u":"/compare/v0#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs v0 by Vercel","h":"Checking the result · v0 by Vercel","x":"The sandbox fixes build errors; the Meta template \"ships with an XR emulator, so you can step into your scene from any browser\" (Quest emulation).","u":"/compare/v0#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs v0 by Vercel","h":"Headsets and devices · Graspable","x":"Immersive web features depend on the device and browser. Preview emulates Quest; test on the target headset. iPhone Tabletop AR uses a tracking mat; native phone AR uses a separate Unity template. Headset sharing uses Wi-Fi or USB. In web projects a Talk button in XR starts a spoken conversation with the agent, in the emulated headset or on a headset sharing the preview over Wi-Fi.","u":"/compare/v0#fact-headsets-and-devices"},{"k":"comparison","t":"Graspable vs v0 by Vercel","h":"Headsets and devices · v0 by Vercel","x":"Meta Quest; a PWA path to the Horizon Store. Vision Pro and Android XR are not mentioned.","u":"/compare/v0#fact-headsets-and-devices"},{"k":"comparison","t":"Graspable vs v0 by Vercel","h":"Your code · Graspable","x":"Editable project files on your disk: React Three Fiber and WebXR for web projects, or the selected native engine’s project. File history and undo are available.","u":"/compare/v0#fact-your-code"},{"k":"comparison","t":"Graspable vs v0 by Vercel","h":"Your code · v0 by Vercel","x":"Download a zip or sync to GitHub, which then \"becomes the source of truth\". Vercel does not own the generated code.","u":"/compare/v0#fact-your-code"},{"k":"comparison","t":"Graspable vs v0 by Vercel","h":"Price · Graspable","x":"Local building with your own supported AI account is free. Monthly plans: Builder $20, Pro $40, Studio $99 per seat with a three-seat minimum. Features and Credits differ by plan (graspable.dev/pricing).","u":"/compare/v0#fact-price"},{"k":"comparison","t":"Graspable vs v0 by Vercel","h":"Price · v0 by Vercel","x":"Free with a daily message limit; Plus $30 per user a month with $30 of credits; Business $100 per user a month; Enterprise custom (v0.app/pricing).","u":"/compare/v0#fact-price"},{"k":"comparison","t":"Graspable vs v0 by Vercel","h":"Native headset apps · Graspable","x":"Pro and Studio include separate native templates for Unity, Unreal Engine and Swift/RealityKit on visionOS. Each uses its engine tools and target-device setup; Graspable does not automatically convert a web or Unity project to Swift.","u":"/compare/v0#fact-native-headset-apps"},{"k":"comparison","t":"Graspable vs v0 by Vercel","h":"Native headset apps · v0 by Vercel","x":"None; web only.","u":"/compare/v0#fact-native-headset-apps"},{"k":"comparison","t":"Graspable vs Lovable","h":"Compare","x":"Both can help build an interactive 3D web app. How do Graspable and Lovable compare when you need headset previews, interaction tests and a place to publish? Lovable is a browser app builder for full-stack web apps, with its own backend, hosting, custom domains and Git sync.","u":"/compare/lovable"},{"k":"comparison","t":"Graspable vs Lovable","h":"What you build · Graspable","x":"A desktop app for Windows, macOS and Linux. Describe a 3D or XR scene, typed or spoken; an agent edits a real project on your computer: React Three Fiber and WebXR, or a native Unity, Unreal Engine or visionOS project.","u":"/compare/lovable#fact-what-you-build"},{"k":"comparison","t":"Graspable vs Lovable","h":"What you build · Lovable","x":"Full-stack web applications. Lovable’s own 3D case study describes a scene with physics and collisions, plus a workflow integrating Three.js code. Its npm guide supports adding libraries by request.","u":"/compare/lovable#fact-what-you-build"},{"k":"comparison","t":"Graspable vs Lovable","h":"The AI · Graspable","x":"Your own AI account: an API key from Anthropic, OpenAI, OpenRouter and others, a ChatGPT subscription, or Claude Code with a Claude subscription. Models come from the provider’s own list.","u":"/compare/lovable#fact-the-ai"},{"k":"comparison","t":"Graspable vs Lovable","h":"The AI · Lovable","x":"Lovable selects the model used for building. Model choice for AI features inside a generated app is a separate capability.","u":"/compare/lovable#fact-the-ai"},{"k":"comparison","t":"Graspable vs Lovable","h":"Where it runs · Graspable","x":"On your computer. Projects, previews, builds, keys and sign-ins stay there. A paid plan can move a project to a Cloud Agent and back.","u":"/compare/lovable#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs Lovable","h":"Where it runs · Lovable","x":"In the browser and Lovable’s cloud: editing, preview, hosting, backend.","u":"/compare/lovable#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs Lovable","h":"Checking the result · Graspable","x":"Web projects must build and load in a browser before a run completes. Native templates use their engine’s checks. Recorded headset tests are replayed when configured.","u":"/compare/lovable#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs Lovable","h":"Checking the result · Lovable","x":"Build mode can inspect errors, test browser workflows and run frontend or backend checks. Most verification tools run when requested. The reviewed pages do not establish headset emulation or recorded XR replay.","u":"/compare/lovable#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs Lovable","h":"Headsets and devices · Graspable","x":"Immersive web features depend on the device and browser. Preview emulates Quest; test on the target headset. iPhone Tabletop AR uses a tracking mat; native phone AR uses a separate Unity template. Headset sharing uses Wi-Fi or USB. In web projects a Talk button in XR starts a spoken conversation with the agent, in the emulated headset or on a headset sharing the preview over Wi-Fi.","u":"/compare/lovable#fact-headsets-and-devices"},{"k":"comparison","t":"Graspable vs Lovable","h":"Headsets and devices · Lovable","x":"The reviewed sources demonstrate browser 3D, not a tested headset or phone AR workflow. WebXR tracking and input still need implementation and checks on the intended device; absence from these docs does not prove they are impossible.","u":"/compare/lovable#fact-headsets-and-devices"},{"k":"comparison","t":"Graspable vs Lovable","h":"Your code · Graspable","x":"Editable project files on your disk: React Three Fiber and WebXR for web projects, or the selected native engine’s project. File history and undo are available.","u":"/compare/lovable#fact-your-code"},{"k":"comparison","t":"Graspable vs Lovable","h":"Your code · Lovable","x":"Two-way GitHub sync is available on all plans, with one active branch at a time. Paid plans offer code downloads; exported projects can be developed and hosted elsewhere.","u":"/compare/lovable#fact-your-code"},{"k":"comparison","t":"Graspable vs Lovable","h":"Price · Graspable","x":"Local building with your own supported AI account is free. Monthly plans: Builder $20, Pro $40, Studio $99 per seat with a three-seat minimum. Features and Credits differ by plan (graspable.dev/pricing).","u":"/compare/lovable#fact-price"},{"k":"comparison","t":"Graspable vs Lovable","h":"Price · Lovable","x":"Free, Pro, Business and Enterprise plans. Check the current subscription page for build allowances and separate Cloud and AI usage charges.","u":"/compare/lovable#fact-price"},{"k":"comparison","t":"Graspable vs Lovable","h":"Native headset apps · Graspable","x":"Pro and Studio include separate native templates for Unity, Unreal Engine and Swift/RealityKit on visionOS. Each uses its engine tools and target-device setup; Graspable does not automatically convert a web or Unity project to Swift.","u":"/compare/lovable#fact-native-headset-apps"},{"k":"comparison","t":"Graspable vs Lovable","h":"Native headset apps · Lovable","x":"Its FAQ distinguishes generated web apps from native mobile projects. It describes PWA or external wrapper routes for phones, not a native Unity, Unreal or visionOS project workflow.","u":"/compare/lovable#fact-native-headset-apps"},{"k":"comparison","t":"Graspable vs Bolt.new","h":"Compare","x":"Bolt can add three.js to a web app on request; Graspable is built around XR. Which one for a 3D scene, and which for an app? Bolt.new, by StackBlitz, is a browser app builder: a chat agent scaffolds and edits a web or Expo mobile project running in the tab, with hosting, a database and GitHub sync.","u":"/compare/bolt"},{"k":"comparison","t":"Graspable vs Bolt.new","h":"What you build · Graspable","x":"A desktop app for Windows, macOS and Linux. Describe a 3D or XR scene, typed or spoken; an agent edits a real project on your computer: React Three Fiber and WebXR, or a native Unity, Unreal Engine or visionOS project.","u":"/compare/bolt#fact-what-you-build"},{"k":"comparison","t":"Graspable vs Bolt.new","h":"What you build · Bolt.new","x":"Websites, web apps, Expo mobile apps (with an Expo Go preview), slides. No WebXR, Quest, Vision Pro or headset preview was found in its docs, release notes or home page.","u":"/compare/bolt#fact-what-you-build"},{"k":"comparison","t":"Graspable vs Bolt.new","h":"The AI · Graspable","x":"Your own AI account: an API key from Anthropic, OpenAI, OpenRouter and others, a ChatGPT subscription, or Claude Code with a Claude subscription. Models come from the provider’s own list.","u":"/compare/bolt#fact-the-ai"},{"k":"comparison","t":"Graspable vs Bolt.new","h":"The AI · Bolt.new","x":"\"Bolt handles model selection behind the scenes\" (Standard and Max agents). No bring-your-own key on bolt.new; the MIT fork bolt.diy supports many providers.","u":"/compare/bolt#fact-the-ai"},{"k":"comparison","t":"Graspable vs Bolt.new","h":"Where it runs · Graspable","x":"On your computer. Projects, previews, builds, keys and sign-ins stay there. A paid plan can move a project to a Cloud Agent and back.","u":"/compare/bolt#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs Bolt.new","h":"Where it runs · Bolt.new","x":"In the browser; Bolt Cloud for hosting, database, auth and storage.","u":"/compare/bolt#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs Bolt.new","h":"Checking the result · Graspable","x":"Web projects must build and load in a browser before a run completes. Native templates use their engine’s checks. Recorded headset tests are replayed when configured.","u":"/compare/bolt#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs Bolt.new","h":"Checking the result · Bolt.new","x":"The agent fixes what it sees in the tab; nothing checks a scene in a headset.","u":"/compare/bolt#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs Bolt.new","h":"Headsets and devices · Graspable","x":"Immersive web features depend on the device and browser. Preview emulates Quest; test on the target headset. iPhone Tabletop AR uses a tracking mat; native phone AR uses a separate Unity template. Headset sharing uses Wi-Fi or USB. In web projects a Talk button in XR starts a spoken conversation with the agent, in the emulated headset or on a headset sharing the preview over Wi-Fi.","u":"/compare/bolt#fact-headsets-and-devices"},{"k":"comparison","t":"Graspable vs Bolt.new","h":"Headsets and devices · Bolt.new","x":"None documented.","u":"/compare/bolt#fact-headsets-and-devices"},{"k":"comparison","t":"Graspable vs Bolt.new","h":"Your code · Graspable","x":"Editable project files on your disk: React Three Fiber and WebXR for web projects, or the selected native engine’s project. File history and undo are available.","u":"/compare/bolt#fact-your-code"},{"k":"comparison","t":"Graspable vs Bolt.new","h":"Your code · Bolt.new","x":"You keep rights to your content; GitHub sync commits every 30 seconds. Uploads are limited to 10 MB on Free and 100 MB on Pro, which matters for 3D assets.","u":"/compare/bolt#fact-your-code"},{"k":"comparison","t":"Graspable vs Bolt.new","h":"Price · Graspable","x":"Local building with your own supported AI account is free. Monthly plans: Builder $20, Pro $40, Studio $99 per seat with a three-seat minimum. Features and Credits differ by plan (graspable.dev/pricing).","u":"/compare/bolt#fact-price"},{"k":"comparison","t":"Graspable vs Bolt.new","h":"Price · Bolt.new","x":"Free with daily tokens; Pro $25 a month for 10M tokens, larger tiers at $50 and $100; Teams $30 a month per member (bolt.new/pricing).","u":"/compare/bolt#fact-price"},{"k":"comparison","t":"Graspable vs Bolt.new","h":"Native headset apps · Graspable","x":"Pro and Studio include separate native templates for Unity, Unreal Engine and Swift/RealityKit on visionOS. Each uses its engine tools and target-device setup; Graspable does not automatically convert a web or Unity project to Swift.","u":"/compare/bolt#fact-native-headset-apps"},{"k":"comparison","t":"Graspable vs Bolt.new","h":"Native headset apps · Bolt.new","x":"None.","u":"/compare/bolt#fact-native-headset-apps"},{"k":"comparison","t":"Graspable vs Cursor","h":"Compare","x":"A developer who knows three.js can build WebXR in Cursor by hand. What does Graspable add, and what does Cursor keep that Graspable does not have? Cursor is an AI code editor with agents that run locally, in the cloud or from a CLI. It knows nothing in particular about XR: building WebXR in Cursor means scaffolding three.js and React Three Fiber yourself and directing the agent.","u":"/compare/cursor"},{"k":"comparison","t":"Graspable vs Cursor","h":"What you build · Graspable","x":"A desktop app for Windows, macOS and Linux. Describe a 3D or XR scene, typed or spoken; an agent edits a real project on your computer: React Three Fiber and WebXR, or a native Unity, Unreal Engine or visionOS project.","u":"/compare/cursor#fact-what-you-build"},{"k":"comparison","t":"Graspable vs Cursor","h":"What you build · Cursor","x":"Anything in a local folder. No XR, three.js or headset notion in its documentation.","u":"/compare/cursor#fact-what-you-build"},{"k":"comparison","t":"Graspable vs Cursor","h":"The AI · Graspable","x":"Your own AI account: an API key from Anthropic, OpenAI, OpenRouter and others, a ChatGPT subscription, or Claude Code with a Claude subscription. Models come from the provider’s own list.","u":"/compare/cursor#fact-the-ai"},{"k":"comparison","t":"Graspable vs Cursor","h":"The AI · Cursor","x":"Cursor’s own pool plus Claude, GPT, Gemini and others; bring-your-own keys for chat models (not for Tab).","u":"/compare/cursor#fact-the-ai"},{"k":"comparison","t":"Graspable vs Cursor","h":"Where it runs · Graspable","x":"On your computer. Projects, previews, builds, keys and sign-ins stay there. A paid plan can move a project to a Cloud Agent and back.","u":"/compare/cursor#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs Cursor","h":"Where it runs · Cursor","x":"Locally (macOS, Windows, Linux); Cloud Agents on paid plans; a CLI.","u":"/compare/cursor#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs Cursor","h":"Checking the result · Graspable","x":"Web projects must build and load in a browser before a run completes. Native templates use their engine’s checks. Recorded headset tests are replayed when configured.","u":"/compare/cursor#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs Cursor","h":"Checking the result · Cursor","x":"A terminal and a desktop browser tool for the agent. No headset emulation, no XR fixtures, no build gate unless you write one.","u":"/compare/cursor#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs Cursor","h":"Headsets and devices · Graspable","x":"Immersive web features depend on the device and browser. Preview emulates Quest; test on the target headset. iPhone Tabletop AR uses a tracking mat; native phone AR uses a separate Unity template. Headset sharing uses Wi-Fi or USB. In web projects a Talk button in XR starts a spoken conversation with the agent, in the emulated headset or on a headset sharing the preview over Wi-Fi.","u":"/compare/cursor#fact-headsets-and-devices"},{"k":"comparison","t":"Graspable vs Cursor","h":"Headsets and devices · Cursor","x":"Whatever you wire up yourself.","u":"/compare/cursor#fact-headsets-and-devices"},{"k":"comparison","t":"Graspable vs Cursor","h":"Your code · Graspable","x":"Editable project files on your disk: React Three Fiber and WebXR for web projects, or the selected native engine’s project. File history and undo are available.","u":"/compare/cursor#fact-your-code"},{"k":"comparison","t":"Graspable vs Cursor","h":"Your code · Cursor","x":"Local files and your own git.","u":"/compare/cursor#fact-your-code"},{"k":"comparison","t":"Graspable vs Cursor","h":"Price · Graspable","x":"Local building with your own supported AI account is free. Monthly plans: Builder $20, Pro $40, Studio $99 per seat with a three-seat minimum. Features and Credits differ by plan (graspable.dev/pricing).","u":"/compare/cursor#fact-price"},{"k":"comparison","t":"Graspable vs Cursor","h":"Price · Cursor","x":"Hobby free; Pro $20 a month; Pro Plus $60; Ultra $200; Teams $40 per user a month (cursor.com/docs/account/pricing).","u":"/compare/cursor#fact-price"},{"k":"comparison","t":"Graspable vs Cursor","h":"Native headset apps · Graspable","x":"Pro and Studio include separate native templates for Unity, Unreal Engine and Swift/RealityKit on visionOS. Each uses its engine tools and target-device setup; Graspable does not automatically convert a web or Unity project to Swift.","u":"/compare/cursor#fact-native-headset-apps"},{"k":"comparison","t":"Graspable vs Cursor","h":"Native headset apps · Cursor","x":"Whatever toolchain you install; Cursor itself has none.","u":"/compare/cursor#fact-native-headset-apps"},{"k":"comparison","t":"Graspable vs Unity","h":"Compare","x":"Unity makes native headset apps. Graspable builds XR with an agent: WebXR scenes that open from a link, and Unity projects too. People compare them because both can end on a Quest. Unity is the game engine most used for standalone headsets: native apps for Meta Quest, Apple Vision Pro, Android XR, phones and consoles, written in C#. Unity AI sits in the editor, and a free Unity CLI and MCP server ","s":"Unity makes native headset apps. Graspable builds XR with an agent: WebXR scenes that open from a link, and Unity projects too. People compare them because both can end on a Quest. Unity is the game engine most used for standalone headsets: native apps for Meta Quest, Apple Vision Pro, Android XR, phones and consoles, written in C#. Unity AI sits in the editor, and a free Unity CLI and MCP server let outside agents drive it.","u":"/compare/unity"},{"k":"comparison","t":"Graspable vs Unity","h":"What you build · Graspable","x":"A desktop app for Windows, macOS and Linux. Describe a 3D or XR scene, typed or spoken; an agent edits a real project on your computer: React Three Fiber and WebXR, or a native Unity, Unreal Engine or visionOS project.","u":"/compare/unity#fact-what-you-build"},{"k":"comparison","t":"Graspable vs Unity","h":"What you build · Unity","x":"Native apps (Quest APKs, visionOS, Android XR, iOS, desktop, consoles) and flat web builds. No official WebXR.","u":"/compare/unity#fact-what-you-build"},{"k":"comparison","t":"Graspable vs Unity","h":"The AI · Graspable","x":"Your own AI account: an API key from Anthropic, OpenAI, OpenRouter and others, a ChatGPT subscription, or Claude Code with a Claude subscription. Models come from the provider’s own list.","u":"/compare/unity#fact-the-ai"},{"k":"comparison","t":"Graspable vs Unity","h":"The AI · Unity","x":"Unity AI (assistant and generators) on Unity-hosted models with metered credits; a free Unity CLI and MCP server for agents such as Claude Code and Codex; Meta XR Operator lets an agent drive a Quest app in the Meta XR Simulator.","u":"/compare/unity#fact-the-ai"},{"k":"comparison","t":"Graspable vs Unity","h":"Where it runs · Graspable","x":"On your computer. Projects, previews, builds, keys and sign-ins stay there. A paid plan can move a project to a Cloud Agent and back.","u":"/compare/unity#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs Unity","h":"Where it runs · Unity","x":"A multi-gigabyte editor on your computer; builds are installed on a device or run in a simulator.","u":"/compare/unity#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs Unity","h":"Checking the result · Graspable","x":"Web projects must build and load in a browser before a run completes. Native templates use their engine’s checks. Recorded headset tests are replayed when configured.","u":"/compare/unity#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs Unity","h":"Checking the result · Unity","x":"Play mode, test frameworks, simulators (Meta XR Simulator, XR Interaction Simulator) and devices; the agent tooling can enter play mode and take screenshots.","u":"/compare/unity#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs Unity","h":"Headsets and devices · Graspable","x":"Immersive web features depend on the device and browser. Preview emulates Quest; test on the target headset. iPhone Tabletop AR uses a tracking mat; native phone AR uses a separate Unity template. Headset sharing uses Wi-Fi or USB. In web projects a Talk button in XR starts a spoken conversation with the agent, in the emulated headset or on a headset sharing the preview over Wi-Fi.","u":"/compare/unity#fact-headsets-and-devices"},{"k":"comparison","t":"Graspable vs Unity","h":"Headsets and devices · Unity","x":"Meta Quest and Meta VR Glasses (Meta XR SDK), Apple Vision Pro (PolySpatial, which needs Unity Pro, Enterprise or Industry), Android XR, phone AR, consoles.","u":"/compare/unity#fact-headsets-and-devices"},{"k":"comparison","t":"Graspable vs Unity","h":"Your code · Graspable","x":"Editable project files on your disk: React Three Fiber and WebXR for web projects, or the selected native engine’s project. File history and undo are available.","u":"/compare/unity#fact-your-code"},{"k":"comparison","t":"Graspable vs Unity","h":"Your code · Unity","x":"A project on your disk; engine source is read-only for Enterprise.","u":"/compare/unity#fact-your-code"},{"k":"comparison","t":"Graspable vs Unity","h":"Price · Graspable","x":"Local building with your own supported AI account is free. Monthly plans: Builder $20, Pro $40, Studio $99 per seat with a three-seat minimum. Features and Credits differ by plan (graspable.dev/pricing).","u":"/compare/unity#fact-price"},{"k":"comparison","t":"Graspable vs Unity","h":"Price · Unity","x":"Personal free under $200K revenue; Pro $2,310 a year per seat; Industry $4,950 a year per seat; Unity AI credits metered, $10 a month on Personal after a trial (unity.com/pricing).","u":"/compare/unity#fact-price"},{"k":"comparison","t":"Graspable vs Unity","h":"Native headset apps · Graspable","x":"Pro and Studio include separate native templates for Unity, Unreal Engine and Swift/RealityKit on visionOS. Each uses its engine tools and target-device setup; Graspable does not automatically convert a web or Unity project to Swift.","u":"/compare/unity#fact-native-headset-apps"},{"k":"comparison","t":"Graspable vs Unity","h":"Native headset apps · Unity","x":"The point of Unity.","u":"/compare/unity#fact-native-headset-apps"},{"k":"comparison","t":"Graspable vs Unreal Engine","h":"Compare","x":"Unreal is for native, high-end real-time graphics; Graspable is for XR built by an agent, mostly as WebXR that opens from a link, and it can work in an Unreal project too. Here is what each gives up. Unreal Engine is Epic’s source-available C++ and Blueprint engine, with OpenXR for headsets and no web export.","u":"/compare/unreal"},{"k":"comparison","t":"Graspable vs Unreal Engine","h":"What you build · Graspable","x":"A desktop app for Windows, macOS and Linux. Describe a 3D or XR scene, typed or spoken; an agent edits a real project on your computer: React Three Fiber and WebXR, or a native Unity, Unreal Engine or visionOS project.","u":"/compare/unreal#fact-what-you-build"},{"k":"comparison","t":"Graspable vs Unreal Engine","h":"What you build · Unreal Engine","x":"Native apps for desktop, consoles, Quest (OpenXR and the Meta plugin), Vision Pro (experimental, Mac only), Android XR (developer preview). No web or WebXR export.","u":"/compare/unreal#fact-what-you-build"},{"k":"comparison","t":"Graspable vs Unreal Engine","h":"The AI · Graspable","x":"Your own AI account: an API key from Anthropic, OpenAI, OpenRouter and others, a ChatGPT subscription, or Claude Code with a Claude subscription. Models come from the provider’s own list.","u":"/compare/unreal#fact-the-ai"},{"k":"comparison","t":"Graspable vs Unreal Engine","h":"The AI · Unreal Engine","x":"An experimental AI Assistant that answers questions and generates snippets (5.7), and an Unreal MCP plugin (5.8) for your own agent.","u":"/compare/unreal#fact-the-ai"},{"k":"comparison","t":"Graspable vs Unreal Engine","h":"Where it runs · Graspable","x":"On your computer. Projects, previews, builds, keys and sign-ins stay there. A paid plan can move a project to a Cloud Agent and back.","u":"/compare/unreal#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs Unreal Engine","h":"Where it runs · Unreal Engine","x":"Tens of gigabytes on your computer; long compiles.","u":"/compare/unreal#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs Unreal Engine","h":"Checking the result · Graspable","x":"Web projects must build and load in a browser before a run completes. Native templates use their engine’s checks. Recorded headset tests are replayed when configured.","u":"/compare/unreal#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs Unreal Engine","h":"Checking the result · Unreal Engine","x":"Play in editor, devices, the Meta XR Simulator (Windows); whatever your agent wires through MCP.","u":"/compare/unreal#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs Unreal Engine","h":"Headsets and devices · Graspable","x":"Immersive web features depend on the device and browser. Preview emulates Quest; test on the target headset. iPhone Tabletop AR uses a tracking mat; native phone AR uses a separate Unity template. Headset sharing uses Wi-Fi or USB. In web projects a Talk button in XR starts a spoken conversation with the agent, in the emulated headset or on a headset sharing the preview over Wi-Fi.","u":"/compare/unreal#fact-headsets-and-devices"},{"k":"comparison","t":"Graspable vs Unreal Engine","h":"Headsets and devices · Unreal Engine","x":"Quest and Meta VR Glasses, Vision Pro (experimental), Android XR (preview), PC VR.","u":"/compare/unreal#fact-headsets-and-devices"},{"k":"comparison","t":"Graspable vs Unreal Engine","h":"Your code · Graspable","x":"Editable project files on your disk: React Three Fiber and WebXR for web projects, or the selected native engine’s project. File history and undo are available.","u":"/compare/unreal#fact-your-code"},{"k":"comparison","t":"Graspable vs Unreal Engine","h":"Your code · Unreal Engine","x":"A project on your disk and engine source for licensees.","u":"/compare/unreal#fact-your-code"},{"k":"comparison","t":"Graspable vs Unreal Engine","h":"Price · Graspable","x":"Local building with your own supported AI account is free. Monthly plans: Builder $20, Pro $40, Studio $99 per seat with a three-seat minimum. Features and Credits differ by plan (graspable.dev/pricing).","u":"/compare/unreal#fact-price"},{"k":"comparison","t":"Graspable vs Unreal Engine","h":"Price · Unreal Engine","x":"Free under $1M revenue, then a 5% royalty above $1M lifetime gross per product; a per-seat licence for non-game commercial use by companies over $1M (unrealengine.com/license).","u":"/compare/unreal#fact-price"},{"k":"comparison","t":"Graspable vs Unreal Engine","h":"Native headset apps · Graspable","x":"Pro and Studio include separate native templates for Unity, Unreal Engine and Swift/RealityKit on visionOS. Each uses its engine tools and target-device setup; Graspable does not automatically convert a web or Unity project to Swift.","u":"/compare/unreal#fact-native-headset-apps"},{"k":"comparison","t":"Graspable vs Unreal Engine","h":"Native headset apps · Unreal Engine","x":"The point of Unreal.","u":"/compare/unreal#fact-native-headset-apps"},{"k":"comparison","t":"Graspable vs PlayCanvas","h":"Compare","x":"Both support coding agents. Do you want PlayCanvas’s collaborative browser editor or Graspable’s local XR projects and built-in run checks? PlayCanvas combines a collaborative browser editor, an open-source web 3D engine and WebXR support. Coding assistants can work with engine projects or connect to the editor through MCP.","u":"/compare/playcanvas"},{"k":"comparison","t":"Graspable vs PlayCanvas","h":"What you build · Graspable","x":"A desktop app for Windows, macOS and Linux. Describe a 3D or XR scene, typed or spoken; an agent edits a real project on your computer: React Three Fiber and WebXR, or a native Unity, Unreal Engine or visionOS project.","u":"/compare/playcanvas#fact-what-you-build"},{"k":"comparison","t":"Graspable vs PlayCanvas","h":"What you build · PlayCanvas","x":"Web 3D apps and games on the PlayCanvas engine, edited in the cloud editor or as code.","u":"/compare/playcanvas#fact-what-you-build"},{"k":"comparison","t":"Graspable vs PlayCanvas","h":"The AI · Graspable","x":"Your own AI account: an API key from Anthropic, OpenAI, OpenRouter and others, a ChatGPT subscription, or Claude Code with a Claude subscription. Models come from the provider’s own list.","u":"/compare/playcanvas#fact-the-ai"},{"k":"comparison","t":"Graspable vs PlayCanvas","h":"The AI · PlayCanvas","x":"Bring a coding assistant. Engine projects have official agent skills; Editor MCP handles scene, asset and script edits. The AI guide also recommends the VS Code extension for editor projects.","u":"/compare/playcanvas#fact-the-ai"},{"k":"comparison","t":"Graspable vs PlayCanvas","h":"Where it runs · Graspable","x":"On your computer. Projects, previews, builds, keys and sign-ins stay there. A paid plan can move a project to a Cloud Agent and back.","u":"/compare/playcanvas#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs PlayCanvas","h":"Where it runs · PlayCanvas","x":"The editor and projects on playcanvas.com; engine-only projects anywhere.","u":"/compare/playcanvas#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs PlayCanvas","h":"Checking the result · Graspable","x":"Web projects must build and load in a browser before a run completes. Native templates use their engine’s checks. Recorded headset tests are replayed when configured.","u":"/compare/playcanvas#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs PlayCanvas","h":"Checking the result · PlayCanvas","x":"Editor MCP can launch an app, capture screenshots, read logs and live state, and send keyboard, mouse or touch input. Your coding assistant can use those tools to verify changes; XR still needs device testing.","u":"/compare/playcanvas#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs PlayCanvas","h":"Headsets and devices · Graspable","x":"Immersive web features depend on the device and browser. Preview emulates Quest; test on the target headset. iPhone Tabletop AR uses a tracking mat; native phone AR uses a separate Unity template. Headset sharing uses Wi-Fi or USB. In web projects a Talk button in XR starts a spoken conversation with the agent, in the emulated headset or on a headset sharing the preview over Wi-Fi.","u":"/compare/playcanvas#fact-headsets-and-devices"},{"k":"comparison","t":"Graspable vs PlayCanvas","h":"Headsets and devices · PlayCanvas","x":"The current platform matrix lists Quest and Pico VR/AR, Android AR, PCVR and Vision Pro VR. Features vary by platform; the matrix does not list iPhone browser AR.","u":"/compare/playcanvas#fact-headsets-and-devices"},{"k":"comparison","t":"Graspable vs PlayCanvas","h":"Your code · Graspable","x":"Editable project files on your disk: React Three Fiber and WebXR for web projects, or the selected native engine’s project. File history and undo are available.","u":"/compare/playcanvas#fact-your-code"},{"k":"comparison","t":"Graspable vs PlayCanvas","h":"Your code · PlayCanvas","x":"Download apps for self-hosting on every plan; projects stored on playcanvas.com unless engine-only. Engine, editor front end and MCP server are MIT.","u":"/compare/playcanvas#fact-your-code"},{"k":"comparison","t":"Graspable vs PlayCanvas","h":"Price · Graspable","x":"Local building with your own supported AI account is free. Monthly plans: Builder $20, Pro $40, Studio $99 per seat with a three-seat minimum. Features and Credits differ by plan (graspable.dev/pricing).","u":"/compare/playcanvas#fact-price"},{"k":"comparison","t":"Graspable vs PlayCanvas","h":"Price · PlayCanvas","x":"Free public projects with 1 GB; Personal $15 per month with 10 GB; Organization $50 per seat per month with 50 GB. Hosting and app downloads for self-hosting are listed on every plan.","u":"/compare/playcanvas#fact-price"},{"k":"comparison","t":"Graspable vs PlayCanvas","h":"Native headset apps · Graspable","x":"Pro and Studio include separate native templates for Unity, Unreal Engine and Swift/RealityKit on visionOS. Each uses its engine tools and target-device setup; Graspable does not automatically convert a web or Unity project to Swift.","u":"/compare/playcanvas#fact-native-headset-apps"},{"k":"comparison","t":"Graspable vs PlayCanvas","h":"Native headset apps · PlayCanvas","x":"Browser-based XR output on the PlayCanvas engine.","u":"/compare/playcanvas#fact-native-headset-apps"},{"k":"comparison","t":"Graspable vs Wonderland Engine","h":"Compare","x":"Both work with coding agents. Choose between Wonderland’s editor and WebAssembly runtime, or Graspable’s local XR project and verification workflow. Wonderland Engine is a desktop editor with a C++ runtime compiled to WebAssembly, made for high-performance WebXR, with an MCP plugin for coding agents.","u":"/compare/wonderland-engine"},{"k":"comparison","t":"Graspable vs Wonderland Engine","h":"What you build · Graspable","x":"A desktop app for Windows, macOS and Linux. Describe a 3D or XR scene, typed or spoken; an agent edits a real project on your computer: React Three Fiber and WebXR, or a native Unity, Unreal Engine or visionOS project.","u":"/compare/wonderland-engine#fact-what-you-build"},{"k":"comparison","t":"Graspable vs Wonderland Engine","h":"What you build · Wonderland Engine","x":"Web 3D and WebXR projects with JavaScript/TypeScript components and a WebAssembly runtime, from a Windows, macOS or Linux editor.","u":"/compare/wonderland-engine#fact-what-you-build"},{"k":"comparison","t":"Graspable vs Wonderland Engine","h":"The AI · Graspable","x":"Your own AI account: an API key from Anthropic, OpenAI, OpenRouter and others, a ChatGPT subscription, or Claude Code with a Claude subscription. Models come from the provider’s own list.","u":"/compare/wonderland-engine#fact-the-ai"},{"k":"comparison","t":"Graspable vs Wonderland Engine","h":"The AI · Wonderland Engine","x":"The official editor MCP plugin connects external coding assistants to scene imports, components and settings. It requires Editor 1.5.0 or later and runs locally while the editor is open.","u":"/compare/wonderland-engine#fact-the-ai"},{"k":"comparison","t":"Graspable vs Wonderland Engine","h":"Where it runs · Graspable","x":"On your computer. Projects, previews, builds, keys and sign-ins stay there. A paid plan can move a project to a Cloud Agent and back.","u":"/compare/wonderland-engine#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs Wonderland Engine","h":"Where it runs · Wonderland Engine","x":"Local editing and static web builds. Wonderland Cloud also offers web hosting and multiplayer services.","u":"/compare/wonderland-engine#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs Wonderland Engine","h":"Checking the result · Graspable","x":"Web projects must build and load in a browser before a run completes. Native templates use their engine’s checks. Recorded headset tests are replayed when configured.","u":"/compare/wonderland-engine#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs Wonderland Engine","h":"Checking the result · Wonderland Engine","x":"The editor packages assets, serves the app and reloads the browser after changes. You choose build and device checks with the connected coding assistant.","u":"/compare/wonderland-engine#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs Wonderland Engine","h":"Headsets and devices · Graspable","x":"Immersive web features depend on the device and browser. Preview emulates Quest; test on the target headset. iPhone Tabletop AR uses a tracking mat; native phone AR uses a separate Unity template. Headset sharing uses Wi-Fi or USB. In web projects a Talk button in XR starts a spoken conversation with the agent, in the emulated headset or on a headset sharing the preview over Wi-Fi.","u":"/compare/wonderland-engine#fact-headsets-and-devices"},{"k":"comparison","t":"Graspable vs Wonderland Engine","h":"Headsets and devices · Wonderland Engine","x":"WebXR on compatible devices and browsers. Its device guide lists Quest, Vision Pro and others, but the table is dated July 2023; verify current features on the intended device. Phone AR can use tracking integrations.","u":"/compare/wonderland-engine#fact-headsets-and-devices"},{"k":"comparison","t":"Graspable vs Wonderland Engine","h":"Your code · Graspable","x":"Editable project files on your disk: React Three Fiber and WebXR for web projects, or the selected native engine’s project. File history and undo are available.","u":"/compare/wonderland-engine#fact-your-code"},{"k":"comparison","t":"Graspable vs Wonderland Engine","h":"Your code · Wonderland Engine","x":"Local project files. The runtime is licensed for distribution inside your product; attribution and release terms apply.","u":"/compare/wonderland-engine#fact-your-code"},{"k":"comparison","t":"Graspable vs Wonderland Engine","h":"Price · Graspable","x":"Local building with your own supported AI account is free. Monthly plans: Builder $20, Pro $40, Studio $99 per seat with a three-seat minimum. Features and Credits differ by plan (graspable.dev/pricing).","u":"/compare/wonderland-engine#fact-price"},{"k":"comparison","t":"Graspable vs Wonderland Engine","h":"Price · Wonderland Engine","x":"Standard is free through $120,000 annual revenue across Wonderland products; a 10% royalty applies to revenue above that threshold. Enterprise seated licences are quoted separately.","u":"/compare/wonderland-engine#fact-price"},{"k":"comparison","t":"Graspable vs Wonderland Engine","h":"Native headset apps · Graspable","x":"Pro and Studio include separate native templates for Unity, Unreal Engine and Swift/RealityKit on visionOS. Each uses its engine tools and target-device setup; Graspable does not automatically convert a web or Unity project to Swift.","u":"/compare/wonderland-engine#fact-native-headset-apps"},{"k":"comparison","t":"Graspable vs Wonderland Engine","h":"Native headset apps · Wonderland Engine","x":"Web output using Wonderland’s browser runtime.","u":"/compare/wonderland-engine#fact-native-headset-apps"},{"k":"comparison","t":"Graspable vs Needle Engine","h":"Compare","x":"Both support coding agents. Do you want Graspable’s local project and run workflow, or Needle’s Unity, Blender and live Inspector workflow? Needle Engine ships 3D to the web from Unity or Blender: the integrations export glTF and a web project on a three.js-based runtime, with hosting on Needle Cloud and an MCP server for agents.","u":"/compare/needle-engine"},{"k":"comparison","t":"Graspable vs Needle Engine","h":"What you build · Graspable","x":"A desktop app for Windows, macOS and Linux. Describe a 3D or XR scene, typed or spoken; an agent edits a real project on your computer: React Three Fiber and WebXR, or a native Unity, Unreal Engine or visionOS project.","u":"/compare/needle-engine#fact-what-you-build"},{"k":"comparison","t":"Graspable vs Needle Engine","h":"What you build · Needle Engine","x":"Web apps exported from Unity or Blender (or written in code) on Needle’s three.js-based runtime; static output you can host anywhere.","u":"/compare/needle-engine#fact-what-you-build"},{"k":"comparison","t":"Graspable vs Needle Engine","h":"The AI · Graspable","x":"Your own AI account: an API key from Anthropic, OpenAI, OpenRouter and others, a ChatGPT subscription, or Claude Code with a Claude subscription. Models come from the provider’s own list.","u":"/compare/needle-engine#fact-the-ai"},{"k":"comparison","t":"Graspable vs Needle Engine","h":"The AI · Needle Engine","x":"Coding assistants have an official skill and local MCP server. Needle Cloud also offers AI chat/project help and paid material and 3D-model generation allowances.","u":"/compare/needle-engine#fact-the-ai"},{"k":"comparison","t":"Graspable vs Needle Engine","h":"Where it runs · Graspable","x":"On your computer. Projects, previews, builds, keys and sign-ins stay there. A paid plan can move a project to a Cloud Agent and back.","u":"/compare/needle-engine#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs Needle Engine","h":"Where it runs · Needle Engine","x":"Locally in Unity or Blender plus a local web project; deploy to Needle Cloud, Netlify, Vercel, GitHub Pages and others.","u":"/compare/needle-engine#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs Needle Engine","h":"Checking the result · Graspable","x":"Web projects must build and load in a browser before a run completes. Native templates use their engine’s checks. Recorded headset tests are replayed when configured.","u":"/compare/needle-engine#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs Needle Engine","h":"Checking the result · Needle Engine","x":"MCP exposes project files, logs and scene information. The live Inspector lets agents inspect and edit objects; an agent can apply manual Inspector changes back to source. Build and device checks belong in that workflow.","u":"/compare/needle-engine#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs Needle Engine","h":"Headsets and devices · Graspable","x":"Immersive web features depend on the device and browser. Preview emulates Quest; test on the target headset. iPhone Tabletop AR uses a tracking mat; native phone AR uses a separate Unity template. Headset sharing uses Wi-Fi or USB. In web projects a Talk button in XR starts a spoken conversation with the agent, in the emulated headset or on a headset sharing the preview over Wi-Fi.","u":"/compare/needle-engine#fact-headsets-and-devices"},{"k":"comparison","t":"Graspable vs Needle Engine","h":"Headsets and devices · Needle Engine","x":"Quest VR/AR, Vision Pro VR and other supported WebXR devices. iOS AR uses an ARKit App Clip; interactive USDZ/QuickLook is a separate path with a subset of functionality, not unrestricted Safari WebXR AR.","u":"/compare/needle-engine#fact-headsets-and-devices"},{"k":"comparison","t":"Graspable vs Needle Engine","h":"Your code · Graspable","x":"Editable project files on your disk: React Three Fiber and WebXR for web projects, or the selected native engine’s project. File history and undo are available.","u":"/compare/needle-engine#fact-your-code"},{"k":"comparison","t":"Graspable vs Needle Engine","h":"Your code · Needle Engine","x":"Web projects and static builds, with a licensed runtime. Hobby requires Needle branding; the pricing table reserves self-hosting and branding removal for Pro and Enterprise.","u":"/compare/needle-engine#fact-your-code"},{"k":"comparison","t":"Graspable vs Needle Engine","h":"Price · Graspable","x":"Local building with your own supported AI account is free. Monthly plans: Builder $20, Pro $40, Studio $99 per seat with a three-seat minimum. Features and Credits differ by plan (graspable.dev/pricing).","u":"/compare/needle-engine#fact-price"},{"k":"comparison","t":"Graspable vs Needle Engine","h":"Price · Needle Engine","x":"Hobby is free for non-commercial use. Pro is €49 per seat per month billed annually (€588), or €99 per seat per month billed quarterly (€297). Enterprise is custom; EDU licences are available by application.","u":"/compare/needle-engine#fact-price"},{"k":"comparison","t":"Graspable vs Needle Engine","h":"Native headset apps · Graspable","x":"Pro and Studio include separate native templates for Unity, Unreal Engine and Swift/RealityKit on visionOS. Each uses its engine tools and target-device setup; Graspable does not automatically convert a web or Unity project to Swift.","u":"/compare/needle-engine#fact-native-headset-apps"},{"k":"comparison","t":"Graspable vs Needle Engine","h":"Native headset apps · Needle Engine","x":"Web deployment, with an additional iOS App Clip AR path.","u":"/compare/needle-engine#fact-native-headset-apps"},{"k":"comparison","t":"Graspable vs Spline","h":"Compare","x":"Both tools let people and agents edit interactive 3D. Compare Spline’s visual scene workflow with Graspable’s editable web and native XR projects. Spline is a visual 3D editor with AI generation, an editing agent and hosted or local MCP connections. Scenes use Spline’s runtime and export options.","u":"/compare/spline"},{"k":"comparison","t":"Graspable vs Spline","h":"What you build · Graspable","x":"A desktop app for Windows, macOS and Linux. Describe a 3D or XR scene, typed or spoken; an agent edits a real project on your computer: React Three Fiber and WebXR, or a native Unity, Unreal Engine or visionOS project.","u":"/compare/spline#fact-what-you-build"},{"k":"comparison","t":"Graspable vs Spline","h":"What you build · Spline","x":"Scenes rendered by Spline’s closed runtime, embedded in sites; code exports to three.js, React and React Three Fiber lose animations and events except for Vanilla JS and React.","u":"/compare/spline#fact-what-you-build"},{"k":"comparison","t":"Graspable vs Spline","h":"The AI · Graspable","x":"Your own AI account: an API key from Anthropic, OpenAI, OpenRouter and others, a ChatGPT subscription, or Claude Code with a Claude subscription. Models come from the provider’s own list.","u":"/compare/spline#fact-the-ai"},{"k":"comparison","t":"Graspable vs Spline","h":"The AI · Spline","x":"An agent inside the editor, 3D generation, and hosted or desktop MCP connections for external agents.","u":"/compare/spline#fact-the-ai"},{"k":"comparison","t":"Graspable vs Spline","h":"Where it runs · Graspable","x":"On your computer. Projects, previews, builds, keys and sign-ins stay there. A paid plan can move a project to a Cloud Agent and back.","u":"/compare/spline#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs Spline","h":"Where it runs · Spline","x":"Browser or desktop app; scenes load from Spline’s CDN by default.","u":"/compare/spline#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs Spline","h":"Checking the result · Graspable","x":"Web projects must build and load in a browser before a run completes. Native templates use their engine’s checks. Recorded headset tests are replayed when configured.","u":"/compare/spline#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs Spline","h":"Checking the result · Spline","x":"The editing agent takes screenshots to check its work. This is different from Graspable’s project build and configured headset tests.","u":"/compare/spline#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs Spline","h":"Headsets and devices · Graspable","x":"Immersive web features depend on the device and browser. Preview emulates Quest; test on the target headset. iPhone Tabletop AR uses a tracking mat; native phone AR uses a separate Unity template. Headset sharing uses Wi-Fi or USB. In web projects a Talk button in XR starts a spoken conversation with the agent, in the emulated headset or on a headset sharing the preview over Wi-Fi.","u":"/compare/spline#fact-headsets-and-devices"},{"k":"comparison","t":"Graspable vs Spline","h":"Headsets and devices · Spline","x":"No WebXR export documented. visionOS through Spline Mirror and native exports in beta; no Quest or Android XR mention.","u":"/compare/spline#fact-headsets-and-devices"},{"k":"comparison","t":"Graspable vs Spline","h":"Your code · Graspable","x":"Editable project files on your disk: React Three Fiber and WebXR for web projects, or the selected native engine’s project. File history and undo are available.","u":"/compare/spline#fact-your-code"},{"k":"comparison","t":"Graspable vs Spline","h":"Your code · Spline","x":"A .spline scene; self-hosted export on the top plan.","u":"/compare/spline#fact-your-code"},{"k":"comparison","t":"Graspable vs Spline","h":"Price · Graspable","x":"Local building with your own supported AI account is free. Monthly plans: Builder $20, Pro $40, Studio $99 per seat with a three-seat minimum. Features and Credits differ by plan (graspable.dev/pricing).","u":"/compare/spline#fact-price"},{"k":"comparison","t":"Graspable vs Spline","h":"Price · Spline","x":"Free with a watermark. Monthly billing per seat: Hobby $15, Pro $30, Max $70. Annual billing equivalents: $12, $25 and $60 per month. Enterprise is custom (spline.design/pricing).","u":"/compare/spline#fact-price"},{"k":"comparison","t":"Graspable vs Spline","h":"Native headset apps · Graspable","x":"Pro and Studio include separate native templates for Unity, Unreal Engine and Swift/RealityKit on visionOS. Each uses its engine tools and target-device setup; Graspable does not automatically convert a web or Unity project to Swift.","u":"/compare/spline#fact-native-headset-apps"},{"k":"comparison","t":"Graspable vs Spline","h":"Native headset apps · Spline","x":"visionOS exports in beta.","u":"/compare/spline#fact-native-headset-apps"},{"k":"comparison","t":"Graspable vs Godot","h":"Compare","x":"Godot is a free engine that can target headsets natively and on the web; Graspable is an agent on a WebXR project. Which one? Godot is a free, MIT-licensed game engine with OpenXR for native headsets (Quest, Android XR and others) and a web export with WebXR. It has no built-in AI, by choice.","u":"/compare/godot"},{"k":"comparison","t":"Graspable vs Godot","h":"What you build · Graspable","x":"A desktop app for Windows, macOS and Linux. Describe a 3D or XR scene, typed or spoken; an agent edits a real project on your computer: React Three Fiber and WebXR, or a native Unity, Unreal Engine or visionOS project.","u":"/compare/godot#fact-what-you-build"},{"k":"comparison","t":"Graspable vs Godot","h":"What you build · Godot","x":"Native apps (desktop, Android including Quest and Android XR, iOS, windowed visionOS) and a web export (WebAssembly, WebGL 2) with WebXR.","u":"/compare/godot#fact-what-you-build"},{"k":"comparison","t":"Graspable vs Godot","h":"The AI · Graspable","x":"Your own AI account: an API key from Anthropic, OpenAI, OpenRouter and others, a ChatGPT subscription, or Claude Code with a Claude subscription. Models come from the provider’s own list.","u":"/compare/godot#fact-the-ai"},{"k":"comparison","t":"Graspable vs Godot","h":"The AI · Godot","x":"None official; third-party plugins and MCP servers only.","u":"/compare/godot#fact-the-ai"},{"k":"comparison","t":"Graspable vs Godot","h":"Where it runs · Graspable","x":"On your computer. Projects, previews, builds, keys and sign-ins stay there. A paid plan can move a project to a Cloud Agent and back.","u":"/compare/godot#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs Godot","h":"Where it runs · Godot","x":"A small editor on your computer; web builds anywhere.","u":"/compare/godot#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs Godot","h":"Checking the result · Graspable","x":"Web projects must build and load in a browser before a run completes. Native templates use their engine’s checks. Recorded headset tests are replayed when configured.","u":"/compare/godot#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs Godot","h":"Checking the result · Godot","x":"Run in the editor or on a device; no agent loop.","u":"/compare/godot#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs Godot","h":"Headsets and devices · Graspable","x":"Immersive web features depend on the device and browser. Preview emulates Quest; test on the target headset. iPhone Tabletop AR uses a tracking mat; native phone AR uses a separate Unity template. Headset sharing uses Wi-Fi or USB. In web projects a Talk button in XR starts a spoken conversation with the agent, in the emulated headset or on a headset sharing the preview over Wi-Fi.","u":"/compare/godot#fact-headsets-and-devices"},{"k":"comparison","t":"Graspable vs Godot","h":"Headsets and devices · Godot","x":"Quest, Pico and Android XR through OpenXR vendor plugins; Steam Frame; windowed visionOS (immersive visionOS merged for a later release); WebXR on the web export.","u":"/compare/godot#fact-headsets-and-devices"},{"k":"comparison","t":"Graspable vs Godot","h":"Your code · Graspable","x":"Editable project files on your disk: React Three Fiber and WebXR for web projects, or the selected native engine’s project. File history and undo are available.","u":"/compare/godot#fact-your-code"},{"k":"comparison","t":"Graspable vs Godot","h":"Your code · Godot","x":"Everything on disk, MIT; the engine’s licence does not apply to what you make.","u":"/compare/godot#fact-your-code"},{"k":"comparison","t":"Graspable vs Godot","h":"Price · Graspable","x":"Local building with your own supported AI account is free. Monthly plans: Builder $20, Pro $40, Studio $99 per seat with a three-seat minimum. Features and Credits differ by plan (graspable.dev/pricing).","u":"/compare/godot#fact-price"},{"k":"comparison","t":"Graspable vs Godot","h":"Price · Godot","x":"$0; donations to the Godot Foundation.","u":"/compare/godot#fact-price"},{"k":"comparison","t":"Graspable vs Godot","h":"Native headset apps · Graspable","x":"Pro and Studio include separate native templates for Unity, Unreal Engine and Swift/RealityKit on visionOS. Each uses its engine tools and target-device setup; Graspable does not automatically convert a web or Unity project to Swift.","u":"/compare/godot#fact-native-headset-apps"},{"k":"comparison","t":"Graspable vs Godot","h":"Native headset apps · Godot","x":"Yes, through OpenXR.","u":"/compare/godot#fact-native-headset-apps"},{"k":"comparison","t":"Graspable vs Zapworks","h":"Compare","x":"Zapworks does phone AR that WebXR alone cannot do on an iPhone; Graspable does headset XR with an agent. Which problem do you have? Zapworks, by Zappar, is a hosted WebAR and WebXR toolkit: a no-code Designer, the Mattercraft code editor with an AI assistant, and computer-vision tracking for phones.","u":"/compare/zapworks"},{"k":"comparison","t":"Graspable vs Zapworks","h":"What you build · Graspable","x":"A desktop app for Windows, macOS and Linux. Describe a 3D or XR scene, typed or spoken; an agent edits a real project on your computer: React Three Fiber and WebXR, or a native Unity, Unreal Engine or visionOS project.","u":"/compare/zapworks#fact-what-you-build"},{"k":"comparison","t":"Graspable vs Zapworks","h":"What you build · Zapworks","x":"WebAR for phones (face, image, world tracking) and WebXR for headsets, in Zappar’s cloud editors; exported as static web code.","u":"/compare/zapworks#fact-what-you-build"},{"k":"comparison","t":"Graspable vs Zapworks","h":"The AI · Graspable","x":"Your own AI account: an API key from Anthropic, OpenAI, OpenRouter and others, a ChatGPT subscription, or Claude Code with a Claude subscription. Models come from the provider’s own list.","u":"/compare/zapworks#fact-the-ai"},{"k":"comparison","t":"Graspable vs Zapworks","h":"The AI · Zapworks","x":"Mattercraft AI Assistant with fixed Claude and Codestral models and monthly credits; no bring-your-own key found.","u":"/compare/zapworks#fact-the-ai"},{"k":"comparison","t":"Graspable vs Zapworks","h":"Where it runs · Graspable","x":"On your computer. Projects, previews, builds, keys and sign-ins stay there. A paid plan can move a project to a Cloud Agent and back.","u":"/compare/zapworks#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs Zapworks","h":"Where it runs · Zapworks","x":"Zappar’s cloud editors and hosting.","u":"/compare/zapworks#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs Zapworks","h":"Checking the result · Graspable","x":"Web projects must build and load in a browser before a run completes. Native templates use their engine’s checks. Recorded headset tests are replayed when configured.","u":"/compare/zapworks#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs Zapworks","h":"Checking the result · Zapworks","x":"Preview in the editor and on devices; no build gate or headset replay.","u":"/compare/zapworks#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs Zapworks","h":"Headsets and devices · Graspable","x":"Immersive web features depend on the device and browser. Preview emulates Quest; test on the target headset. iPhone Tabletop AR uses a tracking mat; native phone AR uses a separate Unity template. Headset sharing uses Wi-Fi or USB. In web projects a Talk button in XR starts a spoken conversation with the agent, in the emulated headset or on a headset sharing the preview over Wi-Fi.","u":"/compare/zapworks#fact-headsets-and-devices"},{"k":"comparison","t":"Graspable vs Zapworks","h":"Headsets and devices · Zapworks","x":"Phones (iOS and Android); Meta Quest, Apple Vision Pro, Magic Leap in Mattercraft.","u":"/compare/zapworks#fact-headsets-and-devices"},{"k":"comparison","t":"Graspable vs Zapworks","h":"Your code · Graspable","x":"Editable project files on your disk: React Three Fiber and WebXR for web projects, or the selected native engine’s project. File history and undo are available.","u":"/compare/zapworks#fact-your-code"},{"k":"comparison","t":"Graspable vs Zapworks","h":"Your code · Zapworks","x":"A zip of static web code; Zappar’s computer-vision libraries stay licensed.","u":"/compare/zapworks#fact-your-code"},{"k":"comparison","t":"Graspable vs Zapworks","h":"Price · Graspable","x":"Local building with your own supported AI account is free. Monthly plans: Builder $20, Pro $40, Studio $99 per seat with a three-seat minimum. Features and Credits differ by plan (graspable.dev/pricing).","u":"/compare/zapworks#fact-price"},{"k":"comparison","t":"Graspable vs Zapworks","h":"Price · Zapworks","x":"Developer $12.99 a month (non-commercial); Pro $315 a month with 12,000 views a year; Enterprise custom (zap.works/pricing).","u":"/compare/zapworks#fact-price"},{"k":"comparison","t":"Graspable vs Zapworks","h":"Native headset apps · Graspable","x":"Pro and Studio include separate native templates for Unity, Unreal Engine and Swift/RealityKit on visionOS. Each uses its engine tools and target-device setup; Graspable does not automatically convert a web or Unity project to Swift.","u":"/compare/zapworks#fact-native-headset-apps"},{"k":"comparison","t":"Graspable vs Zapworks","h":"Native headset apps · Zapworks","x":"None; web.","u":"/compare/zapworks#fact-native-headset-apps"},{"k":"comparison","t":"Graspable vs 8th Wall","h":"Compare","x":"People who built on 8th Wall are looking for what to use next. Graspable is one option for headset XR; here is what it is and is not. 8th Wall was the best-known hosted WebAR platform. Its hosted platform was retired on 28 February 2026; the framework and AR modules are now open source under MIT, and the SLAM engine is a closed binary under a limited-use licence.","u":"/compare/8th-wall"},{"k":"comparison","t":"Graspable vs 8th Wall","h":"What you build · Graspable","x":"A desktop app for Windows, macOS and Linux. Describe a 3D or XR scene, typed or spoken; an agent edits a real project on your computer: React Three Fiber and WebXR, or a native Unity, Unreal Engine or visionOS project.","u":"/compare/8th-wall#fact-what-you-build"},{"k":"comparison","t":"Graspable vs 8th Wall","h":"What you build · 8th Wall","x":"Historically, phone WebAR and Quest WebXR in a hosted editor. Today: a framework and AR modules on GitHub (MIT) you integrate yourself, with three.js, Babylon.js, A-Frame or PlayCanvas.","u":"/compare/8th-wall#fact-what-you-build"},{"k":"comparison","t":"Graspable vs 8th Wall","h":"The AI · Graspable","x":"Your own AI account: an API key from Anthropic, OpenAI, OpenRouter and others, a ChatGPT subscription, or Claude Code with a Claude subscription. Models come from the provider’s own list.","u":"/compare/8th-wall#fact-the-ai"},{"k":"comparison","t":"Graspable vs 8th Wall","h":"The AI · 8th Wall","x":"None.","u":"/compare/8th-wall#fact-the-ai"},{"k":"comparison","t":"Graspable vs 8th Wall","h":"Where it runs · Graspable","x":"On your computer. Projects, previews, builds, keys and sign-ins stay there. A paid plan can move a project to a Cloud Agent and back.","u":"/compare/8th-wall#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs 8th Wall","h":"Where it runs · 8th Wall","x":"Wherever you host it; hosted experiences stay up until 28 February 2027.","u":"/compare/8th-wall#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs 8th Wall","h":"Checking the result · Graspable","x":"Web projects must build and load in a browser before a run completes. Native templates use their engine’s checks. Recorded headset tests are replayed when configured.","u":"/compare/8th-wall#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs 8th Wall","h":"Checking the result · 8th Wall","x":"Yours to set up.","u":"/compare/8th-wall#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs 8th Wall","h":"Headsets and devices · Graspable","x":"Immersive web features depend on the device and browser. Preview emulates Quest; test on the target headset. iPhone Tabletop AR uses a tracking mat; native phone AR uses a separate Unity template. Headset sharing uses Wi-Fi or USB. In web projects a Talk button in XR starts a spoken conversation with the agent, in the emulated headset or on a headset sharing the preview over Wi-Fi.","u":"/compare/8th-wall#fact-headsets-and-devices"},{"k":"comparison","t":"Graspable vs 8th Wall","h":"Headsets and devices · 8th Wall","x":"Phones (world, image, face tracking); Quest WebXR; Vision Pro VR was supported.","u":"/compare/8th-wall#fact-headsets-and-devices"},{"k":"comparison","t":"Graspable vs 8th Wall","h":"Your code · Graspable","x":"Editable project files on your disk: React Three Fiber and WebXR for web projects, or the selected native engine’s project. File history and undo are available.","u":"/compare/8th-wall#fact-your-code"},{"k":"comparison","t":"Graspable vs 8th Wall","h":"Your code · 8th Wall","x":"Yours; the framework is MIT, the SLAM engine binary is under a limited-use licence that allows commercial use inside a broader application.","u":"/compare/8th-wall#fact-your-code"},{"k":"comparison","t":"Graspable vs 8th Wall","h":"Price · Graspable","x":"Local building with your own supported AI account is free. Monthly plans: Builder $20, Pro $40, Studio $99 per seat with a three-seat minimum. Features and Credits differ by plan (graspable.dev/pricing).","u":"/compare/8th-wall#fact-price"},{"k":"comparison","t":"Graspable vs 8th Wall","h":"Price · 8th Wall","x":"Free, no account. Paid subscriptions ended on 28 February 2026.","u":"/compare/8th-wall#fact-price"},{"k":"comparison","t":"Graspable vs 8th Wall","h":"Native headset apps · Graspable","x":"Pro and Studio include separate native templates for Unity, Unreal Engine and Swift/RealityKit on visionOS. Each uses its engine tools and target-device setup; Graspable does not automatically convert a web or Unity project to Swift.","u":"/compare/8th-wall#fact-native-headset-apps"},{"k":"comparison","t":"Graspable vs 8th Wall","h":"Native headset apps · 8th Wall","x":"None.","u":"/compare/8th-wall#fact-native-headset-apps"},{"k":"comparison","t":"Graspable vs Vibe Coding XR (XR Blocks and Gemini)","h":"Compare","x":"Google showed that a chat model can produce a running WebXR app in under a minute. Graspable does the same job as a desktop app with a project, a build gate and a headset emulator. What is the difference? \"Vibe Coding XR\" is a Google Research workflow and paper (March 2026): the Gemini web app, with an XR Blocks Gem, generates a WebXR app on Google’s open-source XR Blocks framework in about a minu","s":"Google showed that a chat model can produce a running WebXR app in under a minute. Graspable does the same job as a desktop app with a project, a build gate and a headset emulator. What is the difference? \"Vibe Coding XR\" is a Google Research workflow and paper (March 2026): the Gemini web app, with an XR Blocks Gem, generates a WebXR app on Google’s open-source XR Blocks framework in about a minute, to run in Chrome’s simulator or on an Android XR headset.","u":"/compare/vibe-coding-xr"},{"k":"comparison","t":"Graspable vs Vibe Coding XR (XR Blocks and Gemini)","h":"What you build · Graspable","x":"A desktop app for Windows, macOS and Linux. Describe a 3D or XR scene, typed or spoken; an agent edits a real project on your computer: React Three Fiber and WebXR, or a native Unity, Unreal Engine or visionOS project.","u":"/compare/vibe-coding-xr#fact-what-you-build"},{"k":"comparison","t":"Graspable vs Vibe Coding XR (XR Blocks and Gemini)","h":"What you build · Vibe Coding XR (XR Blocks and Gemini)","x":"Single HTML, CSS and JavaScript apps on the XR Blocks framework (three.js, Apache-2.0), generated in a Gemini Canvas.","u":"/compare/vibe-coding-xr#fact-what-you-build"},{"k":"comparison","t":"Graspable vs Vibe Coding XR (XR Blocks and Gemini)","h":"The AI · Graspable","x":"Your own AI account: an API key from Anthropic, OpenAI, OpenRouter and others, a ChatGPT subscription, or Claude Code with a Claude subscription. Models come from the provider’s own list.","u":"/compare/vibe-coding-xr#fact-the-ai"},{"k":"comparison","t":"Graspable vs Vibe Coding XR (XR Blocks and Gemini)","h":"The AI · Vibe Coding XR (XR Blocks and Gemini)","x":"Gemini, in the Gemini web app; your Gemini account is the key.","u":"/compare/vibe-coding-xr#fact-the-ai"},{"k":"comparison","t":"Graspable vs Vibe Coding XR (XR Blocks and Gemini)","h":"Where it runs · Graspable","x":"On your computer. Projects, previews, builds, keys and sign-ins stay there. A paid plan can move a project to a Cloud Agent and back.","u":"/compare/vibe-coding-xr#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs Vibe Coding XR (XR Blocks and Gemini)","h":"Where it runs · Vibe Coding XR (XR Blocks and Gemini)","x":"In the chat canvas; the output runs in Chrome’s WebXR simulator or on an Android XR headset.","u":"/compare/vibe-coding-xr#fact-where-it-runs"},{"k":"comparison","t":"Graspable vs Vibe Coding XR (XR Blocks and Gemini)","h":"Checking the result · Graspable","x":"Web projects must build and load in a browser before a run completes. Native templates use their engine’s checks. Recorded headset tests are replayed when configured.","u":"/compare/vibe-coding-xr#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs Vibe Coding XR (XR Blocks and Gemini)","h":"Checking the result · Vibe Coding XR (XR Blocks and Gemini)","x":"None described: no project on disk, no build check, no headset preview pipeline.","u":"/compare/vibe-coding-xr#fact-checking-the-result"},{"k":"comparison","t":"Graspable vs Vibe Coding XR (XR Blocks and Gemini)","h":"Headsets and devices · Graspable","x":"Immersive web features depend on the device and browser. Preview emulates Quest; test on the target headset. iPhone Tabletop AR uses a tracking mat; native phone AR uses a separate Unity template. Headset sharing uses Wi-Fi or USB. In web projects a Talk button in XR starts a spoken conversation with the agent, in the emulated headset or on a headset sharing the preview over Wi-Fi.","u":"/compare/vibe-coding-xr#fact-headsets-and-devices"},{"k":"comparison","t":"Graspable vs Vibe Coding XR (XR Blocks and Gemini)","h":"Headsets and devices · Vibe Coding XR (XR Blocks and Gemini)","x":"Android XR officially (Chrome 136 and later, Galaxy XR, XREAL Aura). Being WebXR, the output can run elsewhere, but Google does not claim that.","u":"/compare/vibe-coding-xr#fact-headsets-and-devices"},{"k":"comparison","t":"Graspable vs Vibe Coding XR (XR Blocks and Gemini)","h":"Your code · Graspable","x":"Editable project files on your disk: React Three Fiber and WebXR for web projects, or the selected native engine’s project. File history and undo are available.","u":"/compare/vibe-coding-xr#fact-your-code"},{"k":"comparison","t":"Graspable vs Vibe Coding XR (XR Blocks and Gemini)","h":"Your code · Vibe Coding XR (XR Blocks and Gemini)","x":"The generated files, copied out of the canvas.","u":"/compare/vibe-coding-xr#fact-your-code"},{"k":"comparison","t":"Graspable vs Vibe Coding XR (XR Blocks and Gemini)","h":"Price · Graspable","x":"Local building with your own supported AI account is free. Monthly plans: Builder $20, Pro $40, Studio $99 per seat with a three-seat minimum. Features and Credits differ by plan (graspable.dev/pricing).","u":"/compare/vibe-coding-xr#fact-price"},{"k":"comparison","t":"Graspable vs Vibe Coding XR (XR Blocks and Gemini)","h":"Price · Vibe Coding XR (XR Blocks and Gemini)","x":"No price beyond a Gemini subscription; the paper and framework are free.","u":"/compare/vibe-coding-xr#fact-price"},{"k":"comparison","t":"Graspable vs Vibe Coding XR (XR Blocks and Gemini)","h":"Native headset apps · Graspable","x":"Pro and Studio include separate native templates for Unity, Unreal Engine and Swift/RealityKit on visionOS. Each uses its engine tools and target-device setup; Graspable does not automatically convert a web or Unity project to Swift.","u":"/compare/vibe-coding-xr#fact-native-headset-apps"},{"k":"comparison","t":"Graspable vs Vibe Coding XR (XR Blocks and Gemini)","h":"Native headset apps · Vibe Coding XR (XR Blocks and Gemini)","x":"None; web.","u":"/compare/vibe-coding-xr#fact-native-headset-apps"},{"k":"pricing","t":"Graspable pricing","x":"Graspable pricing Source: https://graspable.dev/pricing Graspable is available for Windows, macOS and Linux. Local building with your own supported AI account is free. Your AI provider's charges and account limits still apply. Plans add hosted services, templates and publishing capacity. Prices below are in USD, before applicable tax.","u":"/pricing"},{"k":"pricing","t":"Graspable pricing","h":"Explorer","x":"Explorer - Monthly billing: $0 per month. - Annual billing: $0 per year, billed annually. - Included Credits: 0 per month. One-time trial and temporary offers are separate. - Premium templates: not included. - Native XR templates: not included. - Fast mode: not included. - Cloud Agents: available; uses Credits. - Voice requests and spoken answers: available; free minutes every day, then Credits, o","s":"Explorer - Monthly billing: $0 per month. - Annual billing: $0 per year, billed annually. - Included Credits: 0 per month. One-time trial and temporary offers are separate. - Premium templates: not included. - Native XR templates: not included. - Fast mode: not included. - Cloud Agents: available; uses Credits. - Voice requests and spoken answers: available; free minutes every day, then Credits, or your own ElevenLabs key at no Credits. - Published apps: 2; private apps: 0. - Monthly published-app visits: 2000 across the account. - Largest published app: 60 MB. - Cloud project storage: 100 MB per project; 200 MB total. - Credit overage: not offered.","u":"/pricing#plan-explorer"},{"k":"pricing","t":"Graspable pricing","h":"Builder","x":"Builder - Monthly billing: $20 per month. - Annual billing: $192 per year, billed annually. - Included Credits: 150 per month. One-time trial and temporary offers are separate. - Premium templates: included. - Native XR templates: not included. - Fast mode: not included. - Cloud Agents: available; uses Credits. - Voice requests and spoken answers: available; free minutes every day, then Credits, o","s":"Builder - Monthly billing: $20 per month. - Annual billing: $192 per year, billed annually. - Included Credits: 150 per month. One-time trial and temporary offers are separate. - Premium templates: included. - Native XR templates: not included. - Fast mode: not included. - Cloud Agents: available; uses Credits. - Voice requests and spoken answers: available; free minutes every day, then Credits, or your own ElevenLabs key at no Credits. - Published apps: 10; private apps: 3. - Monthly published-app visits: 25000 across the account. - Largest published app: 150 MB. - Cloud project storage: 250 MB per project; 1000 MB total. - Credit overage: $0.20 per Credit, only when the owner opts in.","u":"/pricing#plan-builder"},{"k":"pricing","t":"Graspable pricing","h":"Pro","x":"Pro - Monthly billing: $40 per month. - Annual billing: $384 per year, billed annually. - Included Credits: 250 per month. One-time trial and temporary offers are separate. - Premium templates: included. - Native XR templates: included. - Fast mode: available; uses Credits. - Cloud Agents: available; uses Credits. - Voice requests and spoken answers: available; free minutes every day, then Credits","s":"Pro - Monthly billing: $40 per month. - Annual billing: $384 per year, billed annually. - Included Credits: 250 per month. One-time trial and temporary offers are separate. - Premium templates: included. - Native XR templates: included. - Fast mode: available; uses Credits. - Cloud Agents: available; uses Credits. - Voice requests and spoken answers: available; free minutes every day, then Credits, or your own ElevenLabs key at no Credits. - Published apps: 30; private apps: 10. - Monthly published-app visits: 100000 across the account. - Largest published app: 250 MB. - Cloud project storage: 500 MB per project; 3000 MB total. - Credit overage: $0.18 per Credit, only when the owner opts in.","u":"/pricing#plan-pro"},{"k":"pricing","t":"Graspable pricing","h":"Studio","x":"Studio - Monthly billing: $99 per seat per month. - Annual billing: $948 per seat per year, billed annually. Minimum 3 seats ($297 total per month on monthly billing). - Included Credits: 500 per month per seat, pooled across the team. One-time trial and temporary offers are separate. - Premium templates: included. - Native XR templates: included. - Fast mode: available; uses Credits. - Cloud Agen","s":"Studio - Monthly billing: $99 per seat per month. - Annual billing: $948 per seat per year, billed annually. Minimum 3 seats ($297 total per month on monthly billing). - Included Credits: 500 per month per seat, pooled across the team. One-time trial and temporary offers are separate. - Premium templates: included. - Native XR templates: included. - Fast mode: available; uses Credits. - Cloud Agents: available; uses Credits. - Voice requests and spoken answers: available; free minutes every day, then Credits, or your own ElevenLabs key at no Credits. - Published apps: 100; private apps: 50. - Monthly published-app visits: 500000 across the account. - Largest published app: 500 MB. - Cloud project storage: 1000 MB per project; 10000 MB total. - Credit overage: $0.15 per Credit, only when the owner opts in.","u":"/pricing#plan-studio"}]